last workig state.

.env

@@ -0,0 +1,7 @@
+NEO4J_PASSWORD=mathema2
+LOG_LEVEL=INFO
+LOG_TO_FILE=true
+LOG_FILE=logs/mathema.log
+STATS_DIR=stats/
+LOG_JSON=false
+WEIGHT_PERTURB_PROP=2.0

.idea/neuroevolution.iml

@@ -7,6 +7,9 @@
     <orderEntry type="jdk" jdkName="Python 3.12 (neuroevolution)" jdkType="Python SDK" />
     <orderEntry type="sourceFolder" forTests="false" />
   </component>
+  <component name="PackageRequirementsSettings">
+    <option name="versionSpecifier" value="Don't specify version" />
+  </component>
   <component name="PyDocumentationSettings">
     <option name="format" value="PLAIN" />
     <option name="myDocStringFormat" value="Plain" />

docker-compose.yml

@@ -0,0 +1,18 @@
+services:
+  neo4j:
+    container_name: neo4j
+    image: neo4j:latest
+    ports:
+      - 7474:7474
+      - 7687:7687
+    environment:
+      - NEO4J_AUTH=neo4j/${NEO4J_PASSWORD}
+      - NEO4J_apoc_export_file_enabled=true
+      - NEO4J_apoc_import_file_enabled=true
+      - NEO4J_apoc_import_file_use__neo4j__config=true
+      - NEO4J_PLUGINS=["apoc", "graph-data-science"]
+    volumes:
+      - ./neo4j_db/data:/data
+      - ./neo4j_db/logs:/logs
+      - ./neo4j_db/import:/var/lib/neo4j/import
+      - ./neo4j_db/plugins:/plugins

experiments/car_racing_test/run.py

@@ -6,6 +6,7 @@ obs, info = env.reset()
 for _ in range(1000):
     action = env.action_space.sample()
     _, reward, terminated, truncated, _ = env.step(action)
+
     if terminated or truncated:
         obs, info = env.reset()
 

experiments/phenotype_genotype_map/pehno_geno_map.py

@@ -60,7 +60,7 @@ class Sensor(Actor):
                 if self.sname == "rng":
                     vec = rng_vector(self.vl)
                 else:
-                    # place for own sensors (this will be replaced by scapes down the road)
+                    # place for own sensors (this will be replaced by envs down the road)
                     vec = [0.0] * self.vl
                 # forward an alle Fanouts
                 for pid in self.fanout:

experiments/stochastic_hillclimber/genotype.py

@@ -1,4 +1,4 @@
-# genotype.py
+# genotype.py.old
 import json
 import math
 import random

experiments/stochastic_hillclimber/mtest.py

@@ -1,37 +1,20 @@
 import asyncio
 
 import morphology
-from experiments.stochastic_hillclimber.actors.trainer import Trainer
+from experiments.stochastic_hillclimber.trainer import Trainer
 from genotype import construct, save_genotype
 
 
 def test_genotype_construction():
-    """
-    genotype_data = construct(morphology, hidden_layer_densities=[3, 2])
-
-    # Prüfen, ob Cortex, Sensor, Actuator und Neuronen existieren
-    assert "cortex" in genotype_data
-    assert len(genotype_data["neurons"]) == 3 + 2 + 1  # 3 in 1. HL, 2 in 2. HL, 1 Output
-
-    print("Genotype construction OK")
-    print("Cortex:", genotype_data["cortex"])
-    print("---------------------------------")
-    print("Neurons:", genotype_data["neurons"])
-    print("---------------------------------")
-    print("Actuators:", genotype_data["actuator"])
-
-    save_genotype("test.json", genotype_data)
-    """
-
     trainer = Trainer(
-        morphology_spec=morphology,  # <— wichtig! callable oder "xor_mimic"
-        hidden_layer_densities=[2],  # wie im Buchbeispiel
-        max_attempts=float("inf"),  # MA=inf
-        eval_limit=float("inf"),  # EL=inf
-        fitness_target=99.9,  # FT=99.9
+        morphology_spec=morphology,  
+        hidden_layer_densities=[2],  
+        max_attempts=float("inf"),  
+        eval_limit=float("inf"),  
+        fitness_target=99.9,  
         experimental_file="experimental.json",
         best_file="best.json",
-        exoself_steps_per_eval=0,  # 0 = Scape/Cortex entscheiden über Halt
+        exoself_steps_per_eval=0,  
     )
     asyncio.run(trainer.go())
 

experiments/stochastic_hillclimber/trainer.py

@@ -42,7 +42,7 @@ class Trainer:
         geno = construct(
             self.morphology_spec,
             self.hds,
-            file_name=self.experimental_file,  # <-- schreibt Startnetz nach experimental.json
+            file_name=self.experimental_file,
             add_bias=True
         )
         fitness, evals, cycles, elapsed = await self._evaluate_with_exoself(geno)
@@ -62,7 +62,7 @@ class Trainer:
             print(".........")
 
             if attempt > self.max_attempts or self.eval_acc >= self.eval_limit or self.best_fitness >= self.fitness_target:
-                # Abschlussausgabe wie im Buch
+
                 if self.best_file and os.path.exists(self.best_file):
                     print_genotype(self.best_file)
                 print(
@@ -86,7 +86,6 @@ class Trainer:
             self.cycle_acc += cycles
             self.time_acc += elapsed
 
-            # Besser als bisher?
             if fitness > self.best_fitness:
                 self.best_fitness = fitness
                 if self.best_file and self.experimental_file and os.path.exists(self.experimental_file):
@@ -96,7 +95,6 @@ class Trainer:
                 attempt += 1
 
 
-
 if __name__ == "__main__":
     trainer = Trainer(
         morphology_spec=morphology,

mathema/actors/actuator.py

@@ -1,7 +1,9 @@
-# actors/actuator.py
 import asyncio
+import logging
 
-from actor import Actor
+from mathema.actors.actor import Actor
+
+log = logging.getLogger(__name__)
 
 
 class Actuator(Actor):
@@ -27,7 +29,7 @@ class Actuator(Actor):
                 self.received[from_id] = vec
 
                 if len(self.received) == self.expect:
-                    print("ACTUATOR: collected all signals...")
+                    log.debug("ACTUATOR: collected all signals...")
                     output = []
                     for fid in self.fanin_ids:
                         output.extend(self.received[fid])
@@ -36,19 +38,38 @@ class Actuator(Actor):
                         print(f"Actuator output: {output}")
                         fitness, halt_flag = 1.0, 0
                     elif self.aname == "xor_SendOutput" and self.scape:
-                        print("ACTUATOR: sending action to scape...")
+                        log.debug("ACTUATOR: sending action to scape...")
                         await self.scape.send(("action", output, self))
                         while True:
                             resp = await self.inbox.get()
                             if resp[0] == "result":
-                                print("ACTUATOR: got scape response: ", resp)
+                                log.debug("ACTUATOR: got scape response: %s", resp)
+                                fitness, halt_flag = resp[1], resp[2]
+                                break
+                    elif self.aname == "car_ApplyAction" and self.scape:
+                        y0 = float(output[0]) if len(output) > 0 else 0.0
+                        y1 = float(output[1]) if len(output) > 1 else 0.0
+                        y2 = float(output[2]) if len(output) > 2 else 0.0
+
+                        steer = max(-1.0, min(1.0, y0))
+                        gas = max(0.0, min(1.0, 0.5 * (y1 + 1.0)))
+                        brake = max(0.0, min(1.0, 0.5 * (y2 + 1.0)))
+
+                        action = [steer, gas, brake]
+
+                        log.debug("ACTUATOR: sending action to car scape: %s", action)
+                        await self.scape.send(("action", action, self))
+                        while True:
+                            resp = await self.inbox.get()
+                            if resp[0] == "result":
+                                log.debug("ACTUATOR: got scape response: %s", resp)
                                 fitness, halt_flag = resp[1], resp[2]
                                 break
                     else:
                         fitness, halt_flag = 0.0, 0
 
                     await self.cx_pid.send(("sync", self.aid, fitness, halt_flag))
-                    print("ACTUATOR: sent sync message to cortex.")
+                    log.debug("ACTUATOR: sent sync message to cortex.")
                     self.received.clear()
 
             elif tag == "terminate":

mathema/actors/cortex.py

@@ -1,5 +1,8 @@
 import time
-from actor import Actor
+import logging
+
+from mathema.actors.actor import Actor
+log = logging.getLogger(__name__)
 
 
 class Cortex(Actor):
@@ -19,6 +22,10 @@ class Cortex(Actor):
         self._t0 = None
 
     async def _kick_sensors(self):
+        for n in self.neurons:
+            await n.send(("cycle_start",))
+        for n in self.neurons:
+            await n.send(("tick",))
         for s in self.sensors:
             await s.send(("sync",))
 
@@ -51,15 +58,15 @@ class Cortex(Actor):
                 continue
 
             if tag == "sync" and self.active:
-                print("CORTEX: got sync message: ", msg)
+                log.debug("CORTEX: got sync message: ", msg)
                 _t, aid, fitness, halt_flag = msg
 
-                print("----------------")
-                print("_t:", _t)
-                print("aid:", aid)
-                print("fitness:", fitness)
-                print("halt_flag:", halt_flag)
-                print("----------------")
+                log.debug("----------------")
+                log.debug("_t:", _t)
+                log.debug("aid:", aid)
+                log.debug("fitness:", fitness)
+                log.debug("halt_flag:", halt_flag)
+                log.debug("----------------")
 
                 self.fitness_acc += float(fitness)
                 self.ef_acc += int(halt_flag)
@@ -67,10 +74,10 @@ class Cortex(Actor):
                 if aid in self.awaiting_sync:
                     self.awaiting_sync.remove(aid)
 
-                print("CORTEX: awaiting sync: ", self.awaiting_sync)
+                log.debug("CORTEX: awaiting sync: ", self.awaiting_sync)
 
                 if not self.awaiting_sync:
-                    print("CORTEX: cycle completed.")
+                    log.debug("CORTEX: cycle completed.")
                     self.cycle_acc += 1
 
                     if self.ef_acc > 0:

mathema/actors/neuron.py

@@ -0,0 +1,162 @@
+import math
+import random
+import logging
+from typing import Optional
+
+from mathema.actors.actor import Actor
+
+log = logging.getLogger(__name__)
+
+
+def tanh(x): return math.tanh(x)
+
+
+class Neuron(Actor):
+    def __init__(self, nid, cx_pid, af_name, input_idps, output_pids, bias: Optional[float] = None):
+        super().__init__(f"Neuron-{nid}")
+        self.nid = nid
+        self.cx_pid = cx_pid
+        self.af = tanh if af_name == "tanh" else tanh
+
+        self.inputs = {}
+        self.order = []
+        self._has_recurrent = False
+
+        """
+        for (inp_id, weights) in input_idps:
+            self.order.append(inp_id)
+            self.inputs[inp_id] = {"weights": list(weights), "got": False, "val": None}
+        """
+
+        self.bias = float(bias) if bias is not None else 0.0
+        for inp_id, weights, recurrent in input_idps:
+            recurrent = bool(recurrent)
+            if inp_id == "bias":
+                self.bias = float(weights[0])
+            else:
+                self.order.append(inp_id)
+                self.inputs[inp_id] = {
+                    "weights": list(weights),
+                    "got": False,
+                    "val": [],
+                    "recurrent": recurrent,
+                    "next_val": []
+                }
+                if recurrent:
+                    self._has_recurrent = True
+
+        self._backup_inputs = None
+        self._backup_bias = None
+
+        self.outputs = output_pids
+
+        log.debug(f"Neuron {nid}: inputs={list(self.inputs.keys())}, bias={self.bias}")
+
+    async def run(self):
+        while True:
+            msg = await self.inbox.get()
+            tag = msg[0]
+
+            if tag == "forward":
+                _, from_id, data = msg
+                if from_id not in self.inputs:
+                    continue
+                slot = self.inputs[from_id]
+                if not isinstance(data, list):
+                    data = [float(data)]
+
+                if slot["recurrent"]:
+                    slot["next_val"] = data
+                else:
+                    slot["got"] = True
+                    slot["val"] = data
+
+                if all(self.inputs[i]["got"] for i in self.order):
+                    acc = 0.0
+                    for i in self.order:
+                        w = self.inputs[i]["weights"]
+                        v = self.inputs[i]["val"]
+                        if len(w) != len(v):
+                            raise ValueError(f"Lengths of weights and values must be equal")
+                        acc += sum(wj * vj for wj, vj in zip(w, v))
+                    out = self.af(acc + self.bias)
+                    for pid in self.outputs:
+                        await pid.send(("forward", self.nid, [out]))
+                    for i in self.order:
+                        self.inputs[i]["got"] = False
+                        self.inputs[i]["val"] = []
+                    log.debug(f"Neuron {self.nid}: input_sum={acc + self.bias:.3f}, output={out:.3f}")
+
+            elif tag == "tick":
+
+                if self.order and all(self.inputs[i]["got"] for i in self.order):
+                    acc = 0.0
+                    for i in self.order:
+                        w = self.inputs[i]["weights"]
+                        v = self.inputs[i]["val"]
+                        if len(w) != len(v):
+                            raise ValueError("Lengths of weights and values must be equal")
+                        acc += sum(wj * vj for wj, vj in zip(w, v))
+                    out = self.af(acc + self.bias)
+                    for pid in self.outputs:
+                        await pid.send(("forward", self.nid, [out]))
+                    for i in self.order:
+                        self.inputs[i]["got"] = False
+                        self.inputs[i]["val"] = []
+                    log.debug(f"Neuron {self.nid}: input_sum={acc + self.bias:.3f}, output={out:.3f}")
+
+            elif tag == "get_backup":
+                idps = [(i, self.inputs[i]["weights"]) for i in self.order]
+                idps.append(("bias", [self.bias]))
+                await self.cx_pid.send(("backup_from_neuron", self.nid, idps))
+
+            elif tag == "weight_backup":
+                log.debug(f"Neuron {self.nid}: backing up weights")
+                self._backup_inputs = {k: {"weights": v["weights"][:]} for k, v in self.inputs.items()}
+                self._backup_bias = self.bias
+
+            elif tag == "weight_restore":
+                if self._backup_inputs is not None:
+                    for k in self.inputs:
+                        self.inputs[k]["weights"] = self._backup_inputs[k]["weights"][:]
+                    self.bias = self._backup_bias
+
+            elif tag == "weight_perturb":
+                log.debug(
+                    f"Neuron {self.nid}: perturbing {len([w for i in self.order for w in self.inputs[i]['weights']])}"
+                    f"weights")
+                tot_w = sum(len(self.inputs[i]["weights"]) for i in self.order) + 1
+                mp = 1 / math.sqrt(tot_w)
+                delta_mag = 2.0 * math.pi
+                sat_lim = 2.0 * math.pi
+
+                for i in self.order:
+                    ws = self.inputs[i]["weights"]
+                    for j in range(len(ws)):
+                        if random.random() < mp:
+                            ws[j] = _sat(ws[j] + (random.random() - 0.5) * delta_mag, -sat_lim, sat_lim)
+                if random.random() < mp:
+                    self.bias = _sat(self.bias + (random.random() - 0.5) * delta_mag, -sat_lim, sat_lim)
+
+            elif tag == "cycle_start":
+                for i in self.order:
+                    slot = self.inputs[i]
+                    if slot["recurrent"]:
+                        nv = slot["next_val"]
+                        if not nv:
+                            w = slot["weights"]
+                            nv = [0.0] * max(1, len(w))
+                        slot["val"] = nv
+                        slot["got"] = True
+                        slot["next_val"] = []
+                    else:
+
+                        slot["got"] = False
+                        slot["val"] = []
+
+            elif tag == "terminate":
+                return
+
+
+def _sat(val, lo, hi):
+    return lo if val < lo else (hi if val > hi else val)

mathema/actors/sensor.py

@@ -1,6 +1,8 @@
-# actors/sensor.py
-from actor import Actor
 import random
+import logging
+
+from mathema.actors.actor import Actor
+log = logging.getLogger(__name__)
 
 
 class Sensor(Actor):
@@ -15,15 +17,14 @@ class Sensor(Actor):
 
     async def run(self):
         while True:
-            print("sensor running...")
             msg = await self.inbox.get()
             tag = msg[0]
 
-            print("got sensor message: ", msg)
+            log.debug(f"sensor {self.sid} got sensor message: %s", msg)
 
             if tag == "sync":
                 vec = await self._sense()
-                print("sensed vec: ", vec)
+                log.debug(f"sensor {self.sid} sensed vec: %s", vec)
                 for pid in self.fanout:
                     await pid.send(("forward", self.sid, vec))
 
@@ -34,12 +35,25 @@ class Sensor(Actor):
         if self.sname == "rng":
             return [random.random() for _ in range(self.vl)]
         elif self.sname == "xor_GetInput" and self.scape:
-            print("TODO")
             await self.scape.send(("sense", self.sid, self))
             msg = await self.inbox.get()
             if msg[0] == "percept":
                 return msg[1]
             else:
                 return [0.0] * self.vl
+        elif self.sname == "car_GetFeatures" and self.scape:
+            await self.scape.send(("sense", self.sid, self))
+            msg = await self.inbox.get()
+            if msg[0] == "percept":
+                vec = msg[1]
+
+                out = [max(-1.0, min(1.0, float(x))) for x in vec]
+                if len(out) < self.vl:
+                    out = out + [0.0] * (self.vl - len(out))
+                elif len(out) > self.vl:
+                    out = out[:self.vl]
+                return out
+            else:
+                return [0.0] * self.vl
         else:
             return [0.0] * self.vl

mathema/archive/genotype.py.old

@@ -1,6 +1,5 @@
 # genotype.py
 import json
-import math
 import random
 import time
 from typing import Dict, List, Tuple, Any, Optional

mathema/archive/genotype_new.py.old

@@ -0,0 +1,848 @@
+# genotype.py.old
+from __future__ import annotations
+import asyncio
+import json
+import random
+import time
+from typing import Any, Dict, List, Tuple, Optional, Callable
+
+from mathema.core.db import Neo4jDB
+
+
+# ------------------------------------------------------------
+# ID-Serialisierung im Stil des Buchs (als Strings)
+# ------------------------------------------------------------
+def erlang_id_cortex(uid: float) -> str:
+    return f"{{{{origin,{uid}}},cortex}}"
+
+
+def erlang_id_neuron(layer_idx: int, uid: float) -> str:
+    return f"{{{{{layer_idx},{uid}}},neuron}}"
+
+
+def erlang_id_sensor(uid: float) -> str:
+    return f"{{{uid},sensor}}"
+
+
+def erlang_id_actuator(uid: float) -> str:
+    return f"{{{uid},actuator}}"
+
+
+def now_unique() -> float:
+    return time.time()
+
+
+def generate_ids(n: int) -> List[float]:
+    return [now_unique() + i * 1e-6 for i in range(n)]
+
+
+# ------------------------------------------------------------
+# GenotypeBuilder
+# ------------------------------------------------------------
+class GenotypeBuilder:
+    """
+    Async-Port von Erlangs genotype. Benötigt:
+      - Neo4jDB
+      - morphology.py mit get_InitSensors(morph) / get_InitActuators(morph)
+      - optional: population_monitor mit async create_specie(pop_id, constraint, fingerprint) -> specie_id
+    """
+
+    def __init__(self, db: Neo4jDB, population_monitor: Optional[Any] = None):
+        self.db = db
+        self.population_monitor = population_monitor  # erwartet .create_specie(...)
+
+    # ====================== Public API ======================
+
+    async def construct_Agent(self, specie_id: Any, agent_id: Any, speccon: Dict[str, Any]) -> Dict[str, Any]:
+        random.seed(time.time())
+        generation = 0
+        cx_id, pattern = await self.construct_Cortex(agent_id, generation, speccon)
+
+        agent = {
+            "id": agent_id,
+            "cx_id": cx_id,
+            "specie_id": specie_id,
+            "constraint": speccon,
+            "generation": generation,
+            "pattern": pattern,
+            "evo_hist": [],
+            # default vals
+            "population_id": None,
+            "fingerprint": None,
+            "fitness": None,
+            "innovation_factor": 0
+        }
+        await self._write_agent(agent)
+        await self.update_fingerprint(agent_id)
+        return agent
+
+    async def construct_Cortex(self, agent_id: Any, generation: int, speccon: Dict[str, Any]) -> Tuple[
+        Any, List[Tuple[int, List[Any]]]]:
+        from importlib import import_module
+        morphology_mod = import_module("morphology")
+
+        cx_uid = now_unique()
+        cx_id = erlang_id_cortex(cx_uid)
+        morphology_name = speccon["morphology"]
+
+        init_sensors = morphology_mod.get_InitSensor(morphology_name)
+        init_actuators = morphology_mod.get_InitActuator(morphology_name)
+
+        sensors = []
+        for S in init_sensors:
+            uid = now_unique()
+            sensors.append({
+                **S,
+                "id": erlang_id_sensor(uid),
+                "cx_id": cx_id,
+                "generation": generation,
+                "fanout_ids": S.get("fanout_ids", [])
+            })
+
+        actuators = []
+        for A in init_actuators:
+            uid = now_unique()
+            actuators.append({
+                **A,
+                "id": erlang_id_actuator(uid),
+                "cx_id": cx_id,
+                "generation": generation,
+                "fanin_ids": A.get("fanin_ids", [])
+            })
+
+        neuron_ids = await self.construct_InitialNeuroLayer(cx_id, generation, speccon, sensors, actuators)
+
+        sensor_ids = [s["id"] for s in sensors]
+        actuator_ids = [a["id"] for a in actuators]
+
+        cortex = {
+            "id": cx_id,
+            "agent_id": agent_id,
+            "neuron_ids": neuron_ids,
+            "sensor_ids": sensor_ids,
+            "actuator_ids": actuator_ids
+        }
+        await self._write_cortex_and_io(cortex, sensors, actuators)
+        pattern = [(0, neuron_ids)]
+        return cx_id, pattern
+
+    async def construct_InitialNeuroLayer(
+            self,
+            cx_id: Any,
+            generation: int,
+            speccon: Dict[str, Any],
+            sensors: List[Dict[str, Any]],
+            actuators: List[Dict[str, Any]],
+    ) -> List[Any]:
+        neuron_ids: List[Any] = []
+
+        for A in actuators:
+            vl = int(A["vector_length"])
+            n_uids = generate_ids(vl)
+            n_ids = [erlang_id_neuron(0, u) for u in n_uids]
+
+            for n_id in n_ids:
+                if random.random() >= 0.5:
+                    S = random.choice(sensors)
+                    input_specs = [(S["id"], int(S["vector_length"]))]
+                    S["fanout_ids"] = [n_id] + list(S.get("fanout_ids", []))
+                else:
+                    input_specs = [(S["id"], int(S["vector_length"])) for S in sensors]
+                    for S in sensors:
+                        S["fanout_ids"] = [n_id] + list(S.get("fanout_ids", []))
+
+                await self.construct_Neuron(cx_id, generation, speccon, n_id, input_specs, [A["id"]])
+
+            A["fanin_ids"] = n_ids + list(A.get("fanin_ids", []))
+            neuron_ids.extend(n_ids)
+
+        return neuron_ids
+
+    # Ersetze deine bisherigen _pack_inputs/_unpack_inputs durch diese Versionen:
+
+    """
+    helper functions. because we use list stuff from erlang.
+    this is not needed anymore, if we start to use neo4j friendly
+    data model.
+    """
+
+    def _pack_inputs(self, input_idps: list[dict]) -> tuple[list, list, list]:
+        ids, wflat, lengths = [], [], []
+        for e in input_idps:
+            if "bias" in e:
+                vals = list(e["bias"])
+                ids.append("bias")
+            else:
+                vals = list(e["weights"])
+                ids.append(e["id"])
+            lengths.append(len(vals))
+            wflat.extend(vals)
+        return ids, wflat, lengths
+
+    def _unpack_inputs(self, input_ids: list, input_w: list, input_w_len: list) -> list[dict]:
+        out, i = [], 0
+        for iid, L in zip(input_ids or [], input_w_len or []):
+            chunk = list(input_w[i:i + L]);
+            i += L
+            if iid == "bias":
+                out.append({"bias": chunk})
+            else:
+                out.append({"id": iid, "weights": chunk})
+        return out
+
+    def _normalize_scape(self, v: Any) -> Any:
+        # erlaubt nur primitive Property-Typen
+        if isinstance(v, (str, int, float, bool)) or v is None:
+            return v
+        if isinstance(v, dict) and "private" in v:
+            return v["private"]  # dein bisheriger Fall: {"private":"xor_sim"}
+        return str(v)
+
+    async def construct_Neuron(
+            self,
+            cx_id: Any,
+            generation: int,
+            speccon: Dict[str, Any],
+            n_id: Any,
+            input_specs: List[Tuple[Any, int]],
+            output_ids: List[Any],
+    ) -> None:
+        input_idps = self._create_InputIdPs(input_specs)
+        # expliziter Bias-Eintrag (Projektvorgabe)
+        input_idps.append({"bias": [self._rand_weight()]})
+
+        neuron = {
+            "id": n_id,
+            "generation": generation,
+            "cx_id": cx_id,
+            "af": self._generate_NeuronAF(speccon.get("neural_afs", [])),
+            "input_idps": input_idps,
+            "output_ids": output_ids,
+            "ro_ids": self._calculate_ROIds(n_id, output_ids),
+        }
+        await self._write_neuron(neuron)
+
+    async def update_fingerprint(self, agent_id: Any) -> None:
+        import json
+
+        a = await self._read_agent(agent_id)
+        if not a:
+            return
+
+        cx = await self._read_cortex(a["cx_id"])
+        if not cx:
+            return
+
+        # --- IO lesen ---
+        sensors = await self._read_sensors(cx.get("sensor_ids", []))
+        actuators = await self._read_actuators(cx.get("actuator_ids", []))
+
+        # --- pattern robust aus Agent nehmen (durch _read_agent schon deserialisiert) ---
+        raw_pat = a.get("pattern", []) or []
+        gen_pattern: List[Tuple[int, int]] = []
+        for item in raw_pat:
+            if isinstance(item, (list, tuple)) and len(item) >= 2:
+                li, ids_or_cnt = item[0], item[1]
+                try:
+                    li = int(li)
+                except Exception:
+                    continue
+                if isinstance(ids_or_cnt, (list, tuple)):
+                    cnt = len(ids_or_cnt)
+                elif isinstance(ids_or_cnt, int):
+                    cnt = ids_or_cnt
+                else:
+                    cnt = 0
+                gen_pattern.append((li, cnt))
+
+        # --- Evo-Historie generalisieren (liefert primitive Strukturen) ---
+        gen_evo = self._generalize_EvoHist(a.get("evo_hist", []))
+
+        # --- IO-Deskriptoren auf primitive Typen reduzieren ---
+        def _vec_len(x):
+            return int(x.get("vector_length", x.get("vector_length", 0)) or 0)
+
+        gen_s_desc = [(s.get("name"), _vec_len(s), self._normalize_scape(s.get("scape"))) for s in sensors]
+        gen_a_desc = [(ac.get("name"), _vec_len(ac), self._normalize_scape(ac.get("scape"))) for ac in actuators]
+
+        # --- Fingerprint bauen und als JSON speichern ---
+        fp_obj = {
+            "pattern": gen_pattern,
+            "evo": gen_evo,
+            "sensors": gen_s_desc,
+            "actuators": gen_a_desc,
+        }
+        fp_json = json.dumps(fp_obj, ensure_ascii=False)
+
+        await self._set_agent_fingerprint(agent_id, fp_json)
+
+    # ----------------- Ergänzungen aus Erlang -----------------
+
+    async def speciate(self, agent_id: Any) -> None:
+        """Port von speciate/1."""
+        await self.update_fingerprint(agent_id)
+        A = await self._read_agent(agent_id)
+        if not A:
+            return
+
+        # Test-Agent?
+        if A["id"] == "test":
+            await self._set_agent_fitness(agent_id, None)
+            return
+
+        # Eltern-Specie und Population holen
+        Parent_S = await self._read_specie(A["specie_id"])
+        if not Parent_S:
+            # ohne specie → keine Speziation möglich
+            return
+        P = await self._read_population(Parent_S["population_id"])
+        if not P:
+            return
+
+        # Spezies mit gleichem Fingerprint in der Population finden
+        same: Optional[Any] = await self._find_specie_by_fingerprint(P["id"], A["fingerprint"])
+        if same is None:
+            # Neue Spezies erzeugen (via population_monitor)
+            if not self.population_monitor or not hasattr(self.population_monitor, "create_specie"):
+                raise RuntimeError(
+                    "speciate(): population_monitor.create_specie(pop_id, constraint, fingerprint) fehlt")
+            new_specie_id = await self.population_monitor.create_specie(P["id"], A["constraint"], A["fingerprint"])
+            S = await self._read_specie(new_specie_id)
+            if not S:
+                return
+            # Agent updaten: neue specie_id, fitness=undefined
+            await self._update_agent_fields(agent_id, {"specie_id": new_specie_id, "fitness": None})
+            # Spezies-Agentliste erweitern (prepend wie im Buch reicht, Reihenfolge egal)
+            await self._append_specie_agent(new_specie_id, agent_id, replace=False)
+        else:
+            # Bestehende Spezies updaten
+            await self._update_agent_fields(agent_id, {"specie_id": same, "fitness": None})
+            await self._append_specie_agent(same, agent_id, replace=False)
+
+    async def clone_Agent(self, agent_id: Any, clone_agent_id: Optional[Any] = None) -> Any:
+        """
+        Port von clone_Agent/1,2 (vereinfacht – transaktional durch Reihenfolge).
+        - erzeugt neue IDs nach Buchregeln
+        - kopiert Nodes mit remappten IDs
+        - schreibt Cortex/Agent-Knoten für den Klon
+        """
+        if clone_agent_id is None:
+            clone_agent_id = f"{{{now_unique()},agent}}"
+
+        A = await self._read_agent(agent_id)
+        if not A:
+            raise ValueError(f"agent not found: {agent_id}")
+        Cx = await self._read_cortex(A["cx_id"])
+        if not Cx:
+            raise ValueError(f"cortex not found for agent: {agent_id}")
+
+        # 1) ID-Mapping erzeugen
+        idmap: Dict[str, str] = {}
+        # bias bleibt auf 'bias' → wir mappen nicht (wie im Erlang ETS bias->bias)
+        idmap["bias"] = "bias"
+
+        # Agent
+        idmap[str(agent_id)] = str(clone_agent_id)
+        # Cortex-ID remappen: {{origin,uid},cortex} -> gleicher Layer-Tag 'origin'
+        cx_new = erlang_id_cortex(now_unique())
+        idmap[str(Cx["id"])] = cx_new
+
+        # Neuronen/Sensoren/Aktuatoren IDs remappen
+        for nid in Cx.get("neuron_ids", []):
+            # {{L,uid},neuron} -> {{L,new},neuron}
+            layer = self._parse_layer(nid) or 0
+            idmap[str(nid)] = erlang_id_neuron(layer, now_unique())
+        for sid in Cx.get("sensor_ids", []):
+            idmap[str(sid)] = erlang_id_sensor(now_unique())
+        for aid in Cx.get("actuator_ids", []):
+            idmap[str(aid)] = erlang_id_actuator(now_unique())
+
+        # 2) Originalelemente laden
+        sensors = await self._read_sensors(Cx.get("sensor_ids", []))
+        neurons = await self._read_neurons(Cx.get("neuron_ids", []))
+        actuators = await self._read_actuators(Cx.get("actuator_ids", []))
+
+        # 3) Clones schreiben (Sensors/Actuators/Neurons)
+        #    (IDs, cx_id, *_ids, input_idps remappen)
+        clone_sensors = []
+        for S in sensors:
+            clone_sensors.append({
+                **S,
+                "id": idmap[str(S["id"])],
+                "cx_id": idmap[str(S["cx_id"])],
+                "fanout_ids": [idmap.get(str(x), str(x)) for x in S.get("fanout_ids", [])],
+            })
+        clone_actuators = []
+        for A0 in actuators:
+            clone_actuators.append({
+                **A0,
+                "id": idmap[str(A0["id"])],
+                "cx_id": idmap[str(A0["cx_id"])],
+                "fanin_ids": [idmap.get(str(x), str(x)) for x in A0.get("fanin_ids", [])],
+            })
+        # Neuronen: input_idps, output_ids, ro_ids mappen
+        clone_neurons = []
+        for N in neurons:
+            # Aus flachen Properties rekonstruieren
+            orig_idps = self._unpack_inputs(N.get("input_ids"), N.get("input_w"), N.get("input_w_len"))
+
+            # IDs remappen
+            remapped_idps = []
+            for e in orig_idps:
+                if "bias" in e:
+                    remapped_idps.append({"bias": e["bias"]})
+                else:
+                    remapped_idps.append({"id": idmap.get(str(e["id"]), str(e["id"])), "weights": e["weights"]})
+
+            # Wieder flach packen
+            input_ids, input_w, input_w_len = self._pack_inputs(remapped_idps)
+
+            clone_neurons.append({
+                "id": idmap[str(N["id"])],
+                "cx_id": idmap[str(N["cx_id"])],
+                "generation": N.get("generation", 0),
+                "af": N.get("af", "tanh"),
+                "input_ids": input_ids,
+                "input_w": input_w,
+                "input_w_len": input_w_len,
+                "output_ids": [idmap.get(str(x), str(x)) for x in N.get("output_ids", [])],
+                "ro_ids": [idmap.get(str(x), str(x)) for x in N.get("ro_ids", [])],
+            })
+
+        # 4) Cortex & Agent (Klon) schreiben
+        clone_cortex = {
+            "id": cx_new,
+            "agent_id": idmap[str(agent_id)],
+            "sensor_ids": [c["id"] for c in clone_sensors],
+            "actuator_ids": [c["id"] for c in clone_actuators],
+            "neuron_ids": [c["id"] for c in clone_neurons],
+        }
+        # zuerst Neuronen, dann Cortex + IO, damit HAS_* sofort auflöst
+        for n in clone_neurons:
+            await self._write_neuron(n)
+        await self._write_cortex_and_io(clone_cortex, clone_sensors, clone_actuators)
+
+        # Agent-Klon: minimal id+cx_id updaten; Rest vom Original übernehmen
+        clone_agent = {
+            **A,
+            "id": idmap[str(agent_id)],
+            "cx_id": cx_new,
+        }
+        await self._write_agent(clone_agent)
+        return clone_agent["id"]
+
+    async def test(self) -> None:
+        """Port von test/0 (vereinfacht, ohne Transaktion)."""
+        Specie_Id = "test"
+        Agent_Id = "test"
+        CloneAgent_Id = "test_clone"
+        SpecCon = {"morphology": "xor_mimic", "neural_afs": ["tanh", "cos", "gauss", "abs"]}
+
+        await self.construct_Agent(Specie_Id, Agent_Id, SpecCon)
+        await self.clone_Agent(Agent_Id, CloneAgent_Id)
+        await self.print(Agent_Id)
+        await self.print(CloneAgent_Id)
+        await self.delete_Agent(Agent_Id)
+        await self.delete_Agent(CloneAgent_Id)
+
+    async def create_test(self) -> None:
+        """Port von create_test/0."""
+        Specie_Id = "test"
+        Agent_Id = "test"
+        SpecCon = {"morphology": "xor_mimic", "neural_afs": ["tanh", "cos", "gauss", "abs"]}
+
+        a = await self._read_agent(Agent_Id)
+        if a is None:
+            await self.construct_Agent(Specie_Id, Agent_Id, SpecCon)
+            await self.print(Agent_Id)
+        else:
+            await self.delete_Agent(Agent_Id)
+            await self.construct_Agent(Specie_Id, Agent_Id, SpecCon)
+            await self.print(Agent_Id)
+
+    # ====================== Helper ============================
+
+    def _create_InputIdPs(self, input_specs: List[Tuple[Any, int]]) -> List[Dict[str, Any]]:
+        res: List[Dict[str, Any]] = []
+        for input_id, vl in input_specs:
+            weights = [self._rand_weight() for _ in range(int(vl))]
+            res.append({"id": input_id, "weights": weights})
+        return res
+
+    def _rand_weight(self) -> float:
+        return random.random() - 0.5
+
+    def _generate_NeuronAF(self, afs: List[Any]) -> Any:
+        if not afs:
+            return "tanh"
+        return random.choice(afs)
+
+    def _parse_layer(self, neuron_id: str) -> Optional[int]:
+        try:
+            if "neuron" not in neuron_id:
+                return None
+            inner = neuron_id.split("},neuron")[0]
+            inner = inner.strip("{}").strip("{}")
+            parts = inner.split(",")
+            return int(parts[0])
+        except Exception:
+            return None
+
+    def _calculate_ROIds(self, self_id: Any, output_ids: List[Any]) -> List[Any]:
+        my_layer = self._parse_layer(str(self_id)) or 0
+        acc: List[Any] = []
+        for oid in output_ids:
+            s = str(oid)
+            if "actuator" in s:
+                continue
+            li = self._parse_layer(s)
+            if li is not None and li <= my_layer:
+                acc.append(oid)
+        return acc
+
+    def _generalize_EvoHist(self, evo_hist: List[Any]) -> List[Any]:
+        def strip_id(id_str: Any) -> Any:
+            s = str(id_str)
+            if "neuron" in s:
+                try:
+                    inner = s.split("},neuron")[0].strip("{}").strip("{}")
+                    layer = int(inner.split(",")[0])
+                    return (layer, "neuron")
+                except Exception:
+                    return ("?", "neuron")
+            if "actuator" in s:
+                return ("actuator",)
+            if "sensor" in s:
+                return ("sensor",)
+            return ("?",)
+
+        generalized = []
+        for item in evo_hist:
+            if isinstance(item, (list, tuple)):
+                g = []
+                for el in item:
+                    if isinstance(el, (list, tuple)) and len(el) == 2:
+                        g.append((el[0], strip_id(el[1])))
+                    else:
+                        g.append(strip_id(el))
+                generalized.append(tuple(g))
+            else:
+                generalized.append(item)
+        return generalized
+
+    # =============== Neo4j Schreib-/Lese-Hilfen ===============
+
+    async def _write_agent(self, agent: Dict[str, Any]) -> None:
+        cy = """
+        MERGE (a:agent {id:$id})
+        SET a.generation=$generation,
+            a.population_id=COALESCE($population_id, a.population_id),
+            a.specie_id=$specie_id,
+            a.cx_id=$cx_id,
+            a.fingerprint=COALESCE(a.fingerprint, $fingerprint),
+            a.constraint_json=$constraint_json,
+            a.evo_hist=$evo_hist,
+            a.fitness=COALESCE($fitness, a.fitness),
+            a.innovation_factor=COALESCE($innovation_factor, 0),
+            a.pattern=$pattern_json
+        """
+        payload = {
+            **agent,
+            "constraint_json": json.dumps(agent.get("constraint", {}), ensure_ascii=False),
+            "pattern_json": json.dumps(agent.get("pattern", []), ensure_ascii=False),
+        }
+        await self.db.run_consume(cy, **payload)
+
+    async def _update_agent_fields(self, agent_id: Any, fields: Dict[str, Any]) -> None:
+        sets = ", ".join([f"a.{k}=${k}" for k in fields.keys()])
+        cy = f"MATCH (a:agent {{id:$id}}) SET {sets}"
+        await (await self.db.run_read(cy, id=agent_id, **fields)).consume()
+
+    async def _set_agent_fingerprint(self, agent_id: Any, fingerprint: Any) -> None:
+        cy = "MATCH (a:agent {id:$id}) SET a.fingerprint=$fp"
+        await (await self.db.run_read(cy, id=agent_id, fp=fingerprint)).consume()
+
+    async def _set_agent_fitness(self, agent_id: Any, fitness: Any) -> None:
+        cy = "MATCH (a:agent {id:$id}) SET a.fitness=$fitness"
+        await (await self.db.run_read(cy, id=agent_id, fitness=fitness)).consume()
+
+    async def _read_agent(self, agent_id: Any) -> Optional[Dict[str, Any]]:
+        cy = "MATCH (a:agent {id:$id}) RETURN a"
+        rec = await self.db.read_single(cy, id=agent_id)
+        if not rec:
+            return None
+        a = dict(rec["a"])
+
+        # robuste Deserialisierung
+        import json
+
+        # constraint
+        cj = a.get("constraint_json")
+        if isinstance(cj, str):
+            try:
+                a["constraint"] = json.loads(cj)
+            except Exception:
+                a["constraint"] = {}
+        elif "constraint" not in a or a["constraint"] is None:
+            a["constraint"] = {}
+
+        # pattern – bevorzugt pattern_json; fallback: falls pattern als String gespeichert wurde
+        pj = a.get("pattern_json")
+        if isinstance(pj, str):
+            try:
+                a["pattern"] = json.loads(pj)
+            except Exception:
+                a["pattern"] = []
+        else:
+            p = a.get("pattern")
+            if isinstance(p, str):
+                try:
+                    a["pattern"] = json.loads(p)
+                except Exception:
+                    a["pattern"] = []
+            elif p is None:
+                a["pattern"] = []
+
+        return a
+
+    async def _read_specie(self, specie_id: Any) -> Optional[Dict[str, Any]]:
+        rec = await self.db.read_single("MATCH (s:specie {id:$id}) RETURN s", id=specie_id)
+        return dict(rec["s"]) if rec else None
+
+    async def _read_population(self, population_id: Any) -> Optional[Dict[str, Any]]:
+        rec = await self.db.read_single("MATCH (p:population {id:$id}) RETURN p", id=population_id)
+        return dict(rec["p"]) if rec else None
+
+    async def _find_specie_by_fingerprint(self, population_id: Any, fingerprint: Any) -> Optional[Any]:
+        """Suche in Population eine Spezies mit exakt gleichem Fingerprint (wie im Buch)."""
+        rec = await self.db.read_single("""
+            MATCH (p:population {id:$pid})-[:HAS_SPECIE]->(s:specie)
+            WHERE s.fingerprint = $fp
+            RETURN s.id AS sid
+            LIMIT 1
+        """, pid=population_id, fp=fingerprint)
+        return rec["sid"] if rec else None
+
+    async def _append_specie_agent(self, specie_id: Any, agent_id: Any, replace: bool = False) -> None:
+        """Fügt agent_id in specie.agent_ids ein; erstellt HAS_AGENT-Kante."""
+        cy = """
+        MATCH (s:specie {id:$sid})
+        SET s.agent_ids = CASE WHEN $replace THEN [$aid] ELSE coalesce([$aid] + s.agent_ids, [$aid]) END
+        """
+        await (await self.db.run_read(cy, sid=specie_id, aid=agent_id, replace=replace)).consume()
+        # Beziehung:
+        await (await self.db.run_read("""
+            MATCH (s:specie {id:$sid}), (a:agent {id:$aid})
+            MERGE (s)-[:HAS_AGENT]->(a)
+        """, sid=specie_id, aid=agent_id)).consume()
+
+    async def _write_cortex_and_io(self, cortex: Dict[str, Any], sensors: List[Dict[str, Any]],
+                                   actuators: List[Dict[str, Any]]) -> None:
+
+        sensors = [{**s, "scape": self._normalize_scape(s.get("scape"))} for s in sensors]
+        actuators = [{**a, "scape": self._normalize_scape(a.get("scape"))} for a in actuators]
+
+        await self.db.run_consume("""
+            MERGE (c:cortex {id:$id})
+            SET c.agent_id=$agent_id, c.neuron_ids=$neuron_ids, c.sensor_ids=$sensor_ids, c.actuator_ids=$actuator_ids
+        """, **cortex)
+
+        if sensors:
+            await self.db.run_consume("""
+                UNWIND $sensors AS s
+                MERGE (x:sensor {id:s.id})
+                SET x.name=s.name,
+                    x.cx_id=s.cx_id,
+                    x.scape=s.scape,
+                    x.vl=s.vl,
+                    x.fanout_ids=COALESCE(s.fanout_ids, []),
+                    x.generation=s.generation
+            """, sensors=sensors)
+
+        if actuators:
+            await self.db.run_consume("""
+                UNWIND $actuators AS a
+                MERGE (x:actuator {id:a.id})
+                SET x.name=a.name,
+                    x.cx_id=a.cx_id,
+                    x.scape=a.scape,
+                    x.vl=a.vl,
+                    x.fanin_ids=COALESCE(a.fanin_ids, []),
+                    x.generation=a.generation
+            """, actuators=actuators)
+
+        await self.db.run_consume("""
+            MATCH (c:cortex {id:$cx})
+            WITH c
+            UNWIND c.sensor_ids AS sid
+            MATCH (s:sensor {id:sid})
+            MERGE (c)-[:HAS_SENSOR]->(s)
+        """, cx=cortex["id"])
+
+        await self.db.run_consume("""
+            MATCH (c:cortex {id:$cx})
+            WITH c
+            UNWIND c.actuator_ids AS aid
+            MATCH (a:actuator {id:aid})
+            MERGE (c)-[:HAS_ACTUATOR]->(a)
+        """, cx=cortex["id"])
+
+        await self.db.run_consume("""
+            MATCH (c:cortex {id:$cx})
+            WITH c
+            UNWIND c.neuron_ids AS nid
+            MATCH (n:neuron {id:nid})
+            MERGE (c)-[:HAS_NEURON]->(n)
+        """, cx=cortex["id"])
+
+        await self.db.run_consume("""
+            MATCH (c:cortex {id:$cx})
+            WITH c
+            UNWIND c.sensor_ids AS sid
+            MATCH (s:sensor {id:sid})
+            UNWIND s.fanout_ids AS nid
+            MATCH (n:neuron {id:nid})
+            MERGE (s)-[:FANOUT_TO]->(n)
+        """, cx=cortex["id"])
+
+        await self.db.run_consume("""
+            MATCH (c:cortex {id:$cx})
+            WITH c
+            UNWIND c.actuator_ids AS aid
+            MATCH (a:actuator {id:aid})
+            UNWIND a.fanin_ids AS nid
+            MATCH (n:neuron {id:nid})
+            MERGE (a)-[:FANIN_FROM]->(n)
+        """, cx=cortex["id"])
+
+        # <<< HIER am Ende hinzufügen: OUTPUT_TO-Kanten für alle Neuronen dieses Cortex >>>
+        await self.db.run_consume("""
+            MATCH (c:cortex {id:$cx})
+            UNWIND c.neuron_ids AS nid
+            MATCH (n:neuron {id:nid})
+            UNWIND n.output_ids AS oid
+            CALL {
+              WITH oid
+              OPTIONAL MATCH (m:neuron {id:oid}) RETURN m AS dst
+              UNION
+              WITH oid
+              OPTIONAL MATCH (a:actuator {id:oid}) RETURN a AS dst
+            }
+            WITH n, dst WHERE dst IS NOT NULL
+            MERGE (n)-[:OUTPUT_TO]->(dst)
+        """, cx=cortex["id"])
+
+    async def _write_neuron(self, neuron: Dict[str, Any]) -> None:
+        # Eingabe vereinheitlichen
+        if "input_idps" in neuron:
+            input_ids, input_w, input_w_len = self._pack_inputs(neuron["input_idps"])
+        else:
+            # bereits flach angeliefert
+            input_ids = neuron["input_ids"]
+            input_w = neuron["input_w"]
+            input_w_len = neuron["input_w_len"]
+
+        await self.db.run_consume("""
+            MERGE (n:neuron {id:$id})
+            SET n.generation=$generation,
+                n.cx_id=$cx_id,
+                n.af=$af,
+                n.input_ids=$input_ids,
+                n.input_w=$input_w,
+                n.input_w_len=$input_w_len,
+                n.output_ids=$output_ids,
+                n.ro_ids=$ro_ids
+        """,
+                                  id=neuron["id"],
+                                  generation=neuron["generation"],
+                                  cx_id=neuron["cx_id"],
+                                  af=neuron["af"],
+                                  input_ids=input_ids,
+                                  input_w=input_w,
+                                  input_w_len=input_w_len,
+                                  output_ids=neuron["output_ids"],
+                                  ro_ids=neuron["ro_ids"],
+                                  )
+
+        if neuron["ro_ids"]:
+            await self.db.run_consume("""
+                MATCH (n:neuron {id:$nid})
+                UNWIND $ros AS rid
+                MATCH (src:neuron {id:rid})
+                MERGE (n)-[:READS_OUTPUT_OF]->(src)
+            """, nid=neuron["id"], ros=neuron["ro_ids"])
+
+    async def _read_cortex(self, cx_id: Any) -> Optional[Dict[str, Any]]:
+        rec = await self.db.read_single(
+            "MATCH (c:cortex {id:$id}) RETURN c",
+            id=cx_id,
+        )
+        return dict(rec["c"]) if rec else None
+
+    async def _read_sensors(self, ids: List[Any]) -> List[Dict[str, Any]]:
+        if not ids:
+            return []
+        rows = await self.db.read_all("""
+            UNWIND $ids AS sid
+            MATCH (s:sensor {id:sid}) RETURN s
+        """, ids=ids)
+        return [dict(r["s"]) for r in rows]
+
+    async def _read_actuators(self, ids: List[Any]) -> List[Dict[str, Any]]:
+        if not ids:
+            return []
+        rows = await self.db.read_all("""
+            UNWIND $ids AS aid
+            MATCH (a:actuator {id:aid}) RETURN a
+        """, ids=ids)
+        return [dict(r["a"]) for r in rows]
+
+    async def _read_neurons(self, ids: List[Any]) -> List[Dict[str, Any]]:
+        if not ids:
+            return []
+        rows = await self.db.read_all("""
+            UNWIND $ids AS nid
+            MATCH (n:neuron {id:nid}) RETURN n
+        """, ids=ids)
+        return [dict(r["n"]) for r in rows]
+
+    # -------------- Convenience: delete / print ---------------
+
+    async def delete_Agent(self, agent_id: Any) -> None:
+        a = await self._read_agent(agent_id)
+        if not a:
+            return
+        cx = await self._read_cortex(a["cx_id"])
+        if cx:
+            await (await self.db.run_read("""
+                MATCH (c:cortex {id:$cid})
+                OPTIONAL MATCH (c)-[:HAS_NEURON]->(n:neuron)
+                OPTIONAL MATCH (c)-[:HAS_SENSOR]->(s:sensor)
+                OPTIONAL MATCH (c)-[:HAS_ACTUATOR]->(a:actuator)
+                DETACH DELETE n, s, a, c
+            """, cid=cx["id"])).consume()
+        await (await self.db.run_read("MATCH (a:agent {id:$id}) DETACH DELETE a", id=agent_id)).consume()
+
+    async def print(self, agent_id: Any) -> None:
+        a = await self._read_agent(agent_id)
+        if not a:
+            print("agent not found:", agent_id)
+            return
+        cx = await self._read_cortex(a["cx_id"])
+        print("AGENT:", a)
+        print("CORTEX:", cx)
+        if not cx:
+            return
+        sensors = await self._read_sensors(cx.get("sensor_ids", []))
+        res = await self.db.read_all("""
+            UNWIND $ids AS nid
+            MATCH (n:neuron {id:nid}) RETURN n
+        """, ids=cx.get("neuron_ids", []))
+        neurons = [dict(r["n"]) for r in res]
+        actuators = await self._read_actuators(cx.get("actuator_ids", []))
+        for s in sensors:
+            print("SENSOR:", s)
+        for n in neurons:
+            print("NEURON:", n)
+        for ac in actuators:
+            print("ACTUATOR:", ac)

mathema/car_racing_main.py

@@ -0,0 +1,36 @@
+import asyncio
+from dotenv import load_dotenv
+
+from mathema.core.population_monitor import init_population
+from mathema.utils.logging_config import setup_logging
+
+setup_logging()
+
+import logging
+
+log = logging.getLogger(__name__)
+
+
+async def run_car_test(
+        pop_id: str = "car_pop",
+        gens: int = 200
+):
+    monitor = await init_population((
+        pop_id,
+        [{"morphology": "car_racing_features", "neural_afs": ["tanh"]}],
+        "gt",
+        "competition"
+    ))
+
+    for _ in range(gens):
+        await monitor.gen_ended.wait()
+        s = monitor.state
+        best = await monitor._best_fitness_in_population(s.population_id)
+        log.info(f"[car] gen={s.pop_gen} best_fitness={best:.6f} eval_acc={s.eval_acc}")
+
+    await monitor.stop("normal")
+
+
+if __name__ == "__main__":
+    asyncio.run(run_car_test())
+

mathema/core/db.py

@@ -0,0 +1,66 @@
+from neo4j import AsyncGraphDatabase
+
+NEO4J_CONSTRAINTS = [
+    "CREATE CONSTRAINT cortex_id    IF NOT EXISTS FOR (n:cortex)    REQUIRE n.id IS UNIQUE",
+    "CREATE CONSTRAINT sensor_id    IF NOT EXISTS FOR (n:sensor)    REQUIRE n.id IS UNIQUE",
+    "CREATE CONSTRAINT neuron_id    IF NOT EXISTS FOR (n:neuron)    REQUIRE n.id IS UNIQUE",
+    "CREATE CONSTRAINT actuator_id  IF NOT EXISTS FOR (n:actuator)  REQUIRE n.id IS UNIQUE",
+    "CREATE CONSTRAINT agent_id    IF NOT EXISTS FOR (n:agent)    REQUIRE n.id IS UNIQUE",
+    "CREATE INDEX agent_id IF NOT EXISTS FOR (a:agent) ON (a.id)",
+    "CREATE INDEX agent_population IF NOT EXISTS FOR (a:agent) ON (a.population_id)",
+    "CREATE INDEX cortex_id IF NOT EXISTS FOR (cx:cortex) ON (cx.id)",
+    "CREATE INDEX neuron_id IF NOT EXISTS FOR (n:neuron) ON (n.id)",
+    "CREATE CONSTRAINT spec_id IF NOT EXISTS FOR (s:specie) REQUIRE s.id IS UNIQUE;"
+    "CREATE CONSTRAINT pop_id IF NOT EXISTS FOR (p:population) REQUIRE p.id IS UNIQUE;"
+    # constraint record has no id field, no need to make it unique
+]
+
+
+class Neo4jDB:
+    def __init__(self, uri: str, user: str, password: str, database: str | None = None):
+        self._driver = AsyncGraphDatabase.driver(uri, auth=(user, password))
+        self._database = database
+
+    async def close(self):
+        await self._driver.close()
+
+    """
+    async def run(self, cypher: str, **params):
+        async with self._driver.session(database=self._database) as s:
+            return await s.run(cypher, **params)
+    """
+
+    async def run_read(self, cypher: str, **params):
+        async with self._driver.session(database=self._database) as s:
+            return await s.run(cypher, **params)
+
+    async def read_single(self, cypher: str, **params):
+        async with self._driver.session(database=self._database) as s:
+            res = await s.run(cypher, **params)
+            return await res.single()
+
+    async def read_all(self, cypher: str, **params):
+        async with self._driver.session(database=self._database) as s:
+            res = await s.run(cypher, **params)
+            return [r async for r in res]
+
+    async def run_consume(self, cypher: str, **params):
+        async with self._driver.session(database=self._database) as s:
+            res = await s.run(cypher, **params)
+            return await res.consume()
+
+    async def create_schema(self):
+        async with self._driver.session(database=self._database) as s:
+            for stmt in NEO4J_CONSTRAINTS:
+                await s.run(stmt)
+
+    async def purge_all_nodes(self):
+        async with self._driver.session(database=self._database) as s:
+            await s.run("MATCH (n) DETACH DELETE n")
+
+    async def drop_schema(self):
+        async with self._driver.session(database=self._database) as s:
+            res = await s.run("SHOW CONSTRAINTS")
+            async for record in res:
+                name = record["name"]
+                await s.run(f"DROP CONSTRAINT {name} IF EXISTS")

mathema/core/exoself.py

@@ -2,20 +2,28 @@ import asyncio
 import json
 import math
 import random
+import logging
 from collections import defaultdict
 from typing import Any, Dict, List, Tuple, Optional
 
-from actor import Actor
-from cortex import Cortex
-from sensor import Sensor
-from neuron import Neuron
-from actuator import Actuator
-from scape import XorScape
+from mathema.genotype.neo4j.genotype import load_genotype_snapshot, persist_neuron_backups
+from mathema.actors.actor import Actor
+from mathema.actors.cortex import Cortex
+from mathema.actors.sensor import Sensor
+from mathema.actors.neuron import Neuron
+from mathema.actors.actuator import Actuator
+from mathema.scape.scape import XorScape
+from mathema.scape.car_racing import CarRacingScape
+from mathema.envs.openai_car_racing import CarRacing
+
+log = logging.getLogger(__name__)
 
 
 class Exoself(Actor):
     def __init__(self, genotype: Dict[str, Any], file_name: Optional[str] = None):
         super().__init__("Exoself")
+        self.monitor = None
+        self.agent_id = None
         self.g = genotype
         self.file_name = file_name
 
@@ -26,17 +34,61 @@ class Exoself(Actor):
 
         self.tasks: List[asyncio.Task] = []
 
-        # Training-Stats
         self.highest_fitness = float("-inf")
         self.eval_acc = 0
         self.cycle_acc = 0
         self.time_acc = 0.0
         self.attempt = 0
-        self.MAX_ATTEMPTS = 50
+
+        self.MAX_ATTEMPTS = 10
         self.actuator_scape = None
 
         self._perturbed: List[Neuron] = []
 
+    @classmethod
+    async def start(cls, agent_id: str, monitor) -> "Exoself":
+        try:
+            g = await load_genotype_snapshot(agent_id)
+        except Exception as e:
+
+            log.error(f"[Exoself {agent_id}] START FAILED: {e!r}")
+            try:
+                await monitor.cast(("terminated", agent_id, float("-inf"), 0, 0, 0.0))
+            finally:
+
+                class _Dummy:
+                    async def cast(self, *_a, **_k): pass
+
+                    async def stop(self): pass
+
+                return _Dummy()
+
+        self = cls(g)
+        self.agent_id = agent_id
+        self.monitor = monitor
+
+        async def _runner():
+            fitness = float("-inf")
+            evals = cycles = 0
+            elapsed = 0.0
+            try:
+                fitness, evals, cycles, elapsed = await self.train_until_stop()
+            except Exception as e:
+                log.error(f"[Exoself {self.agent_id}] CRASH in train_until_stop(): {e!r}")
+                fitness = float("-inf")
+                evals = int(self.eval_acc)
+                cycles = int(self.cycle_acc)
+                elapsed = float(self.time_acc)
+            finally:
+                try:
+                    await monitor.cast(("terminated", self.agent_id, fitness, evals, cycles, elapsed))
+                except Exception as e:
+                    log.error(f"[Exoself {self.agent_id}] FAILED to notify monitor: {e!r}")
+
+        loop = asyncio.get_running_loop()
+        self._runner_task = loop.create_task(_runner(), name=f"Exoself-{self.agent_id}")
+        return self
+
     @staticmethod
     def from_file(path: str) -> "Exoself":
         with open(path, "r") as f:
@@ -44,7 +96,7 @@ class Exoself(Actor):
         return Exoself(g, file_name=path)
 
     async def run(self):
-        self._build_pid_map_and_spawn()
+        self.build_pid_map_and_spawn()
 
         self._link_cortex()
 
@@ -64,9 +116,9 @@ class Exoself(Actor):
                 return
 
     async def run_evaluation(self):
-        print("build network and link...")
-        self._build_pid_map_and_spawn()
-        print("link cortex...")
+        log.debug(f"exoself: build network and link...")
+        self.build_pid_map_and_spawn()
+        log.debug(f"exoself: link cortex...")
         self._link_cortex()
 
         for a in self.sensor_actors + self.neuron_actors + self.actuator_actors:
@@ -75,11 +127,9 @@ class Exoself(Actor):
         if self.actuator_scape:
             self.tasks.append(asyncio.create_task(self.actuator_scape.run()))
 
-        print("network actors are running...")
-
         while True:
             msg = await self.inbox.get()
-            print("message in exsoself: ", msg)
+            log.debug("message in exsoself %r: ", msg)
             tag = msg[0]
             if tag == "evaluation_completed":
                 _, fitness, cycles, elapsed = msg
@@ -89,14 +139,8 @@ class Exoself(Actor):
                 await self._terminate_all()
                 return float("-inf"), 0, 0, 0.0
 
-    # ---------- Build ----------
-    def _build_pid_map_and_spawn(self):
-        """
-        Baut Cortex, dann alle Neuronen (mit cx_pid=self.cx_actor), dann verlinkt Outputs schichtweise,
-        dann Sensoren/Aktuatoren (mit cx_pid=self.cx_actor). Achtung: Reihenfolge wichtig.
-        """
+    def build_pid_map_and_spawn(self):
         cx = self.g["cortex"]
-        # Cortex zuerst (damit wir cx_pid an Kinder übergeben können)
         self.cx_actor = Cortex(
             cid=cx["id"],
             exoself_pid=self,
@@ -105,7 +149,8 @@ class Exoself(Actor):
             actuator_pids=[]
         )
 
-        self.actuator_scape = XorScape()
+        env = CarRacing(seed_value=5, render_mode=None)
+        self.actuator_scape = CarRacingScape(env=env)
 
         layers: Dict[int, List[Dict[str, Any]]] = defaultdict(list)
         for n in self.g["neurons"]:
@@ -116,24 +161,35 @@ class Exoself(Actor):
 
         for layer in ordered_layers:
             for n in layer:
-                input_idps = [(iw["input_id"], iw["weights"]) for iw in n["input_weights"]]
+                input_idps = [
+                    (iw["input_id"], iw["weights"], bool(iw.get("recurrent", False)))
+                    for iw in n["input_weights"]
+                ]
                 neuron = Neuron(
                     nid=n["id"],
                     cx_pid=self.cx_actor,
                     af_name=n.get("activation_function", "tanh"),
                     input_idps=input_idps,
-                    output_pids=[]  # füllen wir gleich
+                    output_pids=[],
+                    bias=n.get("bias")
                 )
                 id2neuron_actor[n["id"]] = neuron
                 self.neuron_actors.append(neuron)
 
-        for li in range(len(ordered_layers) - 1):
-            next_pids = [id2neuron_actor[nx["id"]] for nx in ordered_layers[li + 1]]
-            for n in ordered_layers[li]:
-                id2neuron_actor[n["id"]].outputs = next_pids
+        out_map: Dict[Any, set] = {nid: set() for nid in id2neuron_actor.keys()}
+        for layer in ordered_layers:
+            for tgt in layer:
+                tgt_pid = id2neuron_actor[tgt["id"]]
+                for iw in tgt["input_weights"]:
+                    src_id = iw["input_id"]
+                    if src_id in id2neuron_actor:
+                        out_map[src_id].add(tgt_pid)
+        for src_id, targets in out_map.items():
+            id2neuron_actor[src_id].outputs = list(targets)
 
         actuators = self._get_actuators_block()
         if not actuators:
+            log.error(f"genotype does not include 'actuator' or 'actuators' section")
             raise ValueError("Genotype must include 'actuator' or 'actuators'.")
 
         for a in actuators:
@@ -149,26 +205,36 @@ class Exoself(Actor):
             )
             self.actuator_actors.append(actuator)
 
-        if ordered_layers:
-            last_layer = ordered_layers[-1]
-            out_targets = self.actuator_actors
-            for n in last_layer:
-                id2neuron_actor[n["id"]].outputs = out_targets
+        for a in self.actuator_actors:
+            for src_id in a.fanin_ids:
+                assert src_id in id2neuron_actor, f"Actuator {a.aid}: fanin_id {src_id} ist kein Neuron"
+                if src_id in id2neuron_actor:
+                    na = id2neuron_actor[src_id]
+                    if a not in na.outputs:
+                        na.outputs.append(a)
 
         sensors = self._get_sensors_block()
         if not sensors:
+            log.error(f"Genotype must include 'sensor' or 'sensors'.")
             raise ValueError("Genotype must include 'sensor' or 'sensors'.")
 
-        first_layer = ordered_layers[0] if ordered_layers else []
-        first_layer_pids = [id2neuron_actor[n["id"]] for n in first_layer]
+        sensor_targets: Dict[Any, List[Neuron]] = defaultdict(list)
+        for layer in ordered_layers:
+            for tgt in layer:
+                tgt_pid = id2neuron_actor[tgt["id"]]
+                for iw in tgt["input_weights"]:
+                    src_id = iw["input_id"]
+                    if src_id != "bias":
+                        sensor_targets[src_id].append(tgt_pid)
 
         for s in sensors:
+            fanout_pids = sensor_targets.get(s["id"], [])
             sensor = Sensor(
                 sid=s["id"],
                 cx_pid=self.cx_actor,
                 name=s["name"],
                 vector_length=s["vector_length"],
-                fanout_pids=first_layer_pids,
+                fanout_pids=fanout_pids,
                 scape=self.actuator_scape
             )
             self.sensor_actors.append(sensor)
@@ -197,10 +263,9 @@ class Exoself(Actor):
         self.tasks.append(asyncio.create_task(self.cx_actor.run()))
 
     async def train_until_stop(self):
-        self._build_pid_map_and_spawn()
+        self.build_pid_map_and_spawn()
         self._link_cortex()
 
-        # 2) Start tasks
         for a in self.sensor_actors + self.neuron_actors + self.actuator_actors:
             self.tasks.append(asyncio.create_task(a.run()))
         if self.actuator_scape:
@@ -213,9 +278,7 @@ class Exoself(Actor):
             if tag == "evaluation_completed":
                 _, fitness, cycles, elapsed = msg
                 maybe_stats = await self._on_evaluation_completed(fitness, cycles, elapsed)
-                # _on_evaluation_completed() ruft bei Stop bereits _backup_genotype() und _terminate_all()
                 if isinstance(maybe_stats, dict):
-                    # Trainingsende – Daten aus self.* zurückgeben (wie im Buch: Fitness/Evals/Cycles/Time)
                     return (
                         float(self.highest_fitness),
                         int(self.eval_acc),
@@ -232,10 +295,28 @@ class Exoself(Actor):
         self.cycle_acc += int(cycles)
         self.time_acc += float(elapsed)
 
-        print(f"[Exoself] evaluation_completed: fitness={fitness:.6f} cycles={cycles} time={elapsed:.3f}s")
+        log.info(f"[Exoself] evaluation_completed: fitness={fitness:.6f} cycles={cycles} time={elapsed:.3f}s")
+        log.info(f"[Exoself] attempt {self.attempt}")
 
-        REL = 1e-6
-        if fitness > self.highest_fitness * (1.0 + REL):
+        REL = 1e-4
+        ABS_EPS = 1e-2
+
+        valid_fitness = isinstance(fitness, (int, float)) and math.isfinite(fitness)
+        if not valid_fitness:
+            fitness = float("-inf")
+
+        # --- NEW: per-episode log to monitor (only if finite) ---
+        if valid_fitness:
+            try:
+                # fitness is the episodic return from Cortex.fitness_acc
+                await self.monitor.cast(("episode_done", str(self.agent_id), float(fitness), int(self.eval_acc)))
+            except Exception:
+                pass
+
+        thresh = max(ABS_EPS, REL * abs(self.highest_fitness if math.isfinite(self.highest_fitness) else 0.0))
+        improved = fitness > self.highest_fitness + thresh
+        
+        if improved:
             self.highest_fitness = fitness
             self.attempt = 0
             for n in self.neuron_actors:
@@ -246,8 +327,9 @@ class Exoself(Actor):
                 await n.send(("weight_restore",))
 
         if self.attempt >= self.MAX_ATTEMPTS:
-            print(
-                f"[Exoself] STOP. Best fitness={self.highest_fitness:.6f} evals={self.eval_acc} cycles={self.cycle_acc}")
+            log.info(
+                f"[Exoself] STOP. Best fitness={self.highest_fitness:.6f} "
+                f"evals={self.eval_acc} cycles={self.cycle_acc}")
             await self._backup_genotype()
             await self._terminate_all()
             return {
@@ -272,34 +354,31 @@ class Exoself(Actor):
             await n.send(("get_backup",))
 
         backups: List[Tuple[Any, List[Tuple[Any, List[float]]]]] = []
-
         while remaining > 0:
             msg = await self.inbox.get()
             if msg[0] == "backup_from_neuron":
                 _, nid, idps = msg
-                backups.append((nid, idps))
+                backups.append((str(nid), idps))
                 remaining -= 1
 
-        id2n = {n["id"]: n for n in self.g["neurons"]}
+        bias_rows = []
+        edge_rows = []
         for nid, idps in backups:
-            if nid not in id2n:
-                continue
-            new_iw = []
-            bias_val = None
-            for item in idps:
-                if isinstance(item[0], str) and item[0] == "bias":
-                    bias_val = float(item[1]) if not isinstance(item[1], list) else float(item[1][0])
+            for inp_id, weights in idps:
+                if inp_id == "bias":
+                    b = float(weights[0])
+                    bias_rows.append({"nid": str(nid), "bias": b})
                 else:
-                    input_id, weights = item
-                    new_iw.append({"input_id": input_id, "weights": list(weights)})
-            id2n[nid]["input_weights"] = new_iw
-            if bias_val is not None:
-                id2n[nid].setdefault("input_weights", []).append({"input_id": "bias", "weights": [bias_val]})
+                    edge_rows.append({
+                        "from_id": str(inp_id),
+                        "to_id": str(nid),
+                        "weights": [float(x) for x in list(weights)],
+                    })
+
+        await persist_neuron_backups(bias_rows, edge_rows)
 
         if self.file_name:
-            with open(self.file_name, "w") as f:
-                json.dump(self.g, f, indent=2)
-            print(f"[Exoself] Genotype updated → {self.file_name}")
+            log.debug("[Exoself] Hinweis: file_name gesetzt, aber Persistenz läuft über Genotyp-API (Neo4j).")
 
     async def _terminate_all(self):
         for a in self.sensor_actors + self.neuron_actors + self.actuator_actors:

mathema/core/morphology.py

@@ -1,27 +1,30 @@
-# morphology.py
-import time
+import uuid
+import logging
 from typing import Any, Callable, Dict, List, Union
 
+log = logging.getLogger(__name__)
+
 MorphologyType = Union[str, Callable[[str], List[Dict[str, Any]]]]
 
 
-def generate_id() -> float:
-    now = time.time()
-    return 1.0 / now
+def generate_id() -> str:
+    return uuid.uuid4().hex
 
 
-def get_InitSensor(morphology: MorphologyType) -> Dict[str, Any]:
+def get_InitSensor(morphology: MorphologyType):
     sensors = get_Sensors(morphology)
     if not sensors:
+        log.error("Morphology has no sensors.")
         raise ValueError("Morphology has no sensors.")
-    return sensors[0]
+    return [sensors[0]]
 
 
-def get_InitActuator(morphology: MorphologyType) -> Dict[str, Any]:
+def get_InitActuator(morphology: MorphologyType):
     actuators = get_Actuators(morphology)
     if not actuators:
+        log.error("Morphology has no actuators.")
         raise ValueError("Morphology has no actuators.")
-    return actuators[0]
+    return [actuators[0]]
 
 
 def get_Sensors(morphology: MorphologyType) -> List[Dict[str, Any]]:
@@ -38,22 +41,24 @@ def _resolve_morphology(morphology: MorphologyType) -> Callable[[str], List[Dict
     if callable(morphology):
         return morphology
 
-    # 2) String -> Registry
     if isinstance(morphology, str):
         reg = {
             "xor_mimic": xor_mimic,
+            "car_racing_features": car_racing_features
         }
         if morphology in reg:
             return reg[morphology]
+        log.error(f"Unknown morphology name: {morphology}")
         raise ValueError(f"Unknown morphology name: {morphology}")
 
     try:
-        # Ist es ein Modul mit einer Funktion 'xor_mimic'?
+
         if hasattr(morphology, "xor_mimic") and callable(getattr(morphology, "xor_mimic")):
             return getattr(morphology, "xor_mimic")
     except Exception:
         pass
 
+    log.error("morphology must be a callable, a module with 'xor_mimic', or a registered string key")
     raise TypeError("morphology must be a callable, a module with 'xor_mimic', or a registered string key")
 
 
@@ -61,20 +66,47 @@ def xor_mimic(kind: str) -> List[Dict[str, Any]]:
     if kind == "sensors":
         return [
             {
-                "id": generate_id(),
                 "name": "xor_GetInput",
                 "vector_length": 2,
-                "scape": {"private": "xor_sim"}
+                "scape": "xor_sim"
             }
         ]
     elif kind == "actuators":
         return [
             {
-                "id": generate_id(),
                 "name": "xor_SendOutput",
                 "vector_length": 1,
-                "scape": {"private": "xor_sim"}
+                "scape": "xor_sim"
             }
         ]
     else:
+        log.error(f"xor_mimic: unsupported kind '{kind}', expected 'sensors' or 'actuators'")
         raise ValueError(f"xor_mimic: unsupported kind '{kind}', expected 'sensors' or 'actuators'")
+
+
+def car_racing_features(kind: str) -> List[Dict[str, Any]]:
+    """
+    car racing morphology
+    """
+    LOOK_AHEAD = 10
+    feature_len = LOOK_AHEAD + 6
+
+    if kind == "sensors":
+        return [
+            {
+                "name": "car_GetFeatures",
+                "vector_length": feature_len,
+                "scape": "car_racing"
+            }
+        ]
+    elif kind == "actuators":
+        return [
+            {
+                "name": "car_ApplyAction",
+                "vector_length": 3,
+                "scape": "car_racing"
+            }
+        ]
+    else:
+        log.error(f"car_racing_features: unsupported kind '{kind}', expected 'sensors' or 'actuators'")
+        raise ValueError("car_racing_features: kind must be 'sensors' or 'actuators'")

mathema/core/polis.py

@@ -0,0 +1,172 @@
+from __future__ import annotations
+import asyncio
+import logging
+from dataclasses import dataclass, field
+from typing import Any, Callable, Dict, List, Optional
+
+from mathema.core.db import Neo4jDB
+from mathema.actors.actor import Actor
+
+log = logging.getLogger(__name__)
+
+
+# ===============================================================
+# Datentypen
+# ===============================================================
+
+@dataclass
+class ScapeEntry:
+    type: str
+    actor: Actor  # Actor-Instanz (Pid-Äquivalent)
+    task: asyncio.Task  # laufender run()-Task
+    parameters: dict = field(default_factory=dict)
+
+
+@dataclass
+class PolisState:
+    active_mods: List[str] = field(default_factory=list)
+    active_scapes: List[ScapeEntry] = field(default_factory=list)
+
+
+# ===============================================================
+# Polis-Klasse
+# ===============================================================
+
+class Polis:
+    """
+    Python-Asyncio Port von Erlangs polis.
+    - start/stop, create/reset wie im Buch
+    - Mods = Nebenmodule mit start/stop
+    - Scapes = Actor-basierte Environments (z.B. XorScape)
+    - Neo4j statt Mnesia
+    """
+
+    def __init__(
+            self,
+            mods: List[tuple[str, Callable[[], Any], Callable[[], Any]]] = None,
+            public_scapes: List[tuple[str, dict, Callable[[], Actor]]] = None,
+            neo4j_uri: str = "bolt://localhost:7687",
+            neo4j_user: str = "neo4j",
+            neo4j_pass: str = "mathema2",
+            neo4j_db: Optional[str] = None,
+    ):
+        self._mods = mods or []  # (name, async_start, async_stop)
+        self._public_scapes = public_scapes or []  # (type, params, factory->Actor)
+        self._state = PolisState()
+        self._db = Neo4jDB(neo4j_uri, neo4j_user, neo4j_pass, neo4j_db)
+        self._lock = asyncio.Lock()
+        self._online_evt = asyncio.Event()
+        self._closing = False
+        self._scapes: Dict[str, ScapeEntry] = {}  # type -> ScapeEntry
+
+    # -----------------------------------------------------------
+    # API
+    # -----------------------------------------------------------
+
+    async def sync(self) -> None:
+        """Im Erlang-Code: make:all([load]). Hier als Platzhalter (Hot-Reload)."""
+        return
+
+    async def create(self) -> None:
+        """Schema in Neo4j anlegen (Constraints/Indizes)."""
+        await self._db.create_schema()
+
+    async def reset(self) -> None:
+        """Alle Daten löschen, Schema bleibt erhalten."""
+        await self._db.purge_all_nodes()
+        await self._db.create_schema()
+
+    async def start(self) -> None:
+        async with self._lock:
+            if self._online_evt.is_set():
+                log.info("polis already online.")
+                return
+
+            await self._db.create_schema()
+            # await self._start_supmods(self._mods)
+            active_scapes = await self._start_scapes(self._public_scapes)
+
+            self._state = PolisState(
+                active_mods=[name for (name, _s, _t) in self._mods],
+                active_scapes=active_scapes,
+            )
+            self._online_evt.set()
+            log.info("******** Polis: ##MATHEMA## is now online.")
+
+    async def stop(self, reason: str = "normal") -> None:
+        async with self._lock:
+            if not self._online_evt.is_set():
+                log.info("polis is offline")
+                return
+
+            self._closing = True
+            await self._stop_scapes(self._state.active_scapes)
+            # await self._stop_supmods(self._mods)
+            await self._db.close()
+            self._online_evt.clear()
+            log.info(f"******** Polis: ##MATHEMA## is now offline, terminated with reason:{reason}")
+
+    async def get_scape(self, scape_type: str) -> Optional[Actor]:
+        """Pid-Äquivalent (Actor-Instanz) für Scape-Typ zurückgeben."""
+        entry = self._scapes.get(scape_type)
+        return entry.actor if entry else None
+
+    # -----------------------------------------------------------
+    # intern: Mods
+    # -----------------------------------------------------------
+
+    """
+
+    async def _start_supmods(self, mods: List[tuple[str, Callable[[], Any], Callable[[], Any]]]) -> None:
+        for name, start_fn, _stop_fn in mods:
+            print(f"Starting mod: {name}")
+            res = start_fn()
+            if asyncio.iscoroutine(res):
+                await res
+
+    async def _stop_supmods(self, mods: List[tuple[str, Callable[[], Any], Callable[[], Any]]]) -> None:
+        for name, _start_fn, stop_fn in mods:
+            print(f"Stopping mod: {name}")
+            res = stop_fn()
+            if asyncio.iscoroutine(res):
+                await res
+
+    """
+
+    # -----------------------------------------------------------
+    # intern: Scapes
+    # -----------------------------------------------------------
+
+    async def _start_scapes(
+            self, scapes_cfg: List[tuple[str, dict, Callable[[], Actor]]]
+    ) -> List[ScapeEntry]:
+        active: List[ScapeEntry] = []
+        for scape_type, params, factory in scapes_cfg:
+            actor: Actor = factory()  # Actor-Instanz erzeugen
+            task = asyncio.create_task(actor.run(), name=f"scape:{scape_type}")
+            entry = ScapeEntry(type=scape_type, actor=actor, task=task, parameters=params or {})
+            self._scapes[scape_type] = entry
+            active.append(entry)
+            log.debug(f"Scape started: {scape_type} -> task={task.get_name()}")
+        return active
+
+    async def _stop_scapes(self, scapes: List[ScapeEntry]) -> None:
+        for e in scapes:
+            try:
+                await e.actor.send(("terminate",))
+            except Exception as ex:
+                log.warning(f"Warn: terminate send failed for {e.type}: {ex}")
+
+        for e in scapes:
+            try:
+                await asyncio.wait_for(e.task, timeout=2.0)
+            except asyncio.TimeoutError:
+                e.task.cancel()
+                try:
+                    await e.task
+                except asyncio.CancelledError:
+                    pass
+            except Exception as ex:
+                log.warning(f"Warn: scape {e.type} exit error: {ex}")
+
+        self._scapes.clear()

mathema/core/population_monitor.py

@@ -0,0 +1,658 @@
+"""
+module used to monitor a population of neural networks
+solving a shared or isolated scape.
+"""
+
+from __future__ import annotations
+import asyncio
+import json
+import math
+import os
+import random
+import uuid
+import logging
+from dataclasses import dataclass, field
+from typing import Any, List, Literal, Optional, Sequence, Tuple, Callable, Awaitable
+
+from mathema.genotype.neo4j.genotype import (
+    neo4j,
+    construct_agent,
+    clone_agent,
+    delete_agent,
+    update_fingerprint,
+)
+from mathema.genotype.neo4j.genotype_mutator import GenotypeMutator
+from mathema.utils import stats
+from mathema.core.exoself import Exoself
+
+log = logging.getLogger(__name__)
+
+OpTag = Literal["continue", "pause", "done"]
+SelectionAlgorithm = Literal["competition", "top3"]
+
+EFF: float = 0.05
+SURVIVAL_PERCENTAGE: float = 0.5
+SPECIE_SIZE_LIMIT: int = 10
+INIT_SPECIE_SIZE: int = 10
+
+GENERATION_LIMIT: int = 1000
+EVALUATIONS_LIMIT: int = 100_000
+FITNESS_GOAL: float = 6000
+
+INIT_POPULATION_ID: str = "test"
+INIT_OP_MODE: str = "gt"
+INIT_SELECTION_ALGO: SelectionAlgorithm = "competition"
+INIT_CONSTRAINTS: list[dict] = [
+    {"morphology": "xor_mimic", "neural_afs": ["tanh"]},
+]
+
+EXOSELF_START: Optional[Callable[[str, "PopulationMonitor"], Awaitable[Any]]] = None
+
+
+def _new_id() -> str:
+    return uuid.uuid4().hex
+
+
+def _mean(xs: Sequence[float]) -> float:
+    return sum(xs) / len(xs) if xs else 0.0
+
+
+def _std(xs: Sequence[float]) -> float:
+    if not xs: return 0.0
+    m = _mean(xs)
+    return math.sqrt(sum((x - m) ** 2 for x in xs) / len(xs))
+
+
+async def _read_all(q: str, **params):
+    return await neo4j.read_all(q, **params)
+
+
+async def _run(q: str, **params):
+    return await neo4j.run_consume(q, **params)
+
+
+@dataclass
+class MonitorState:
+    op_mode: str
+    population_id: str
+    selection_algorithm: SelectionAlgorithm
+    op_tag: OpTag = "continue"
+
+    active: List[Tuple[str, Any]] = field(default_factory=list)
+    agent_ids: List[str] = field(default_factory=list)
+
+    tot_agents: int = 0
+    agents_left: int = 0
+
+    pop_gen: int = 0
+    eval_acc: int = 0
+    cycle_acc: int = 0
+    time_acc: int = 0
+    rows: List[dict] = field(default_factory=list)
+
+
+async def _population_aggregate(population_id: str) -> dict:
+    rows = await _read_all("""
+                MATCH (a:agent {population_id:$pid})
+                RETURN collect(coalesce(toFloat(a.fitness),0.0)) AS fs
+            """, pid=str(population_id))
+    fs = [float(x) for x in (rows[0]["fs"] if rows else [])]
+
+    if not fs:
+        return {"cum_fitness": 0.0, "avg": 0.0, "std": 0.0, "best": 0.0, "min": 0.0, "n": 0, "agents": 0}
+    n = len(fs)
+    s = sum(fs)
+    m = s / n
+    var = sum((x - m) ** 2 for x in fs) / n
+
+    return {
+        "cum_fitness": s,
+        "avg": m,
+        "std": math.sqrt(var),
+        "best": max(fs),
+        "min": min(fs),
+        "n": n,
+        "agents": n
+    }
+
+
+class PopulationMonitor:
+    """
+    Orchestriert Generationen: spawn -> warten -> selektieren/mutieren -> next.
+    Expects exoself.start(agent_id, monitor) and exoself will cast(("terminated", aid, fitness, eval_acc, cycle_acc, time_acc)).
+    """
+
+    def __init__(self, op_mode: str, population_id: str, selection_algorithm: SelectionAlgorithm):
+        self.state = MonitorState(op_mode, population_id, selection_algorithm)
+        self.inbox: asyncio.Queue = asyncio.Queue()
+        self._task: Optional[asyncio.Task] = None
+        self._stopped_evt = asyncio.Event()
+        self.mutator = GenotypeMutator(neo4j)
+        self.gen_ended = asyncio.Event()
+        self._exoself_start = EXOSELF_START or Exoself.start
+
+        # logging stuff
+        self.run_id = None
+        self._t0 = None
+        self._best_so_far = float("-inf")
+        self.train_time_sec = 30*60
+
+        # logging file handles
+        self._episodes_f = None
+        self._progress_f = None
+
+    @classmethod
+    async def start(cls, op_mode: str, population_id: str,
+                    selection_algorithm: SelectionAlgorithm) -> "PopulationMonitor":
+        self = cls(op_mode, population_id, selection_algorithm)
+        self._task = asyncio.create_task(self._run(), name=f"PopulationMonitor-{population_id}")
+        stats.register_atexit(population_id, lambda: list(self.state.rows))
+
+        # init logging
+        loop = asyncio.get_running_loop()
+        self._t0 = loop.time()
+        self.run_id = f"{population_id}__{uuid.uuid4().hex[:8]}"
+        os.makedirs(f"runs/{self.run_id}", exist_ok=True)
+
+        self._episodes_f = open(f"runs/{self.run_id}/episodes.csv", "w", buffering=1)
+        self._episodes_f.write("t_sec,pop_gen,agent_id,eval_idx,episode_return\n")
+
+        self._progress_f = open(f"runs/{self.run_id}/progress.csv", "w", buffering=1)
+        self._progress_f.write("t_sec,t_norm,pop_gen,best_gen,best_so_far,avg,std\n")
+
+        self._deadline_task = asyncio.create_task(self._stop_after_deadline(), name=f"Deadline-{self.run_id}")
+
+        await self._init_generation()
+
+        return self
+
+    async def _stop_after_deadline(self):
+        await asyncio.sleep(self.train_time_sec)
+        await self.inbox.put(("stop", "normal"))
+
+    async def cast(self, msg: tuple) -> None:
+        await self.inbox.put(msg)
+
+    async def stop(self, mode: Literal["normal", "shutdown"] = "normal") -> None:
+        await self.inbox.put(("stop", mode))
+        await self._stopped_evt.wait()
+
+    async def _run(self) -> None:
+        try:
+            while True:
+                msg = await self.inbox.get()
+                tag = msg[0] if isinstance(msg, tuple) else None
+
+                if tag == "stop":
+                    await self._handle_stop(msg[1])
+                    break
+                elif tag == "pause":
+                    if self.state.op_tag == "continue":
+                        self.state.op_tag = "pause"
+                        log.debug("Population Monitor will pause after this generation.")
+                elif tag == "continue":
+                    if self.state.op_tag == "pause":
+                        self.state.op_tag = "continue"
+                        await self._init_generation()
+                elif tag == "episode_done":
+                    await self._handle_episode_done(*msg[1:])
+                elif tag == "terminated":
+                    await self._handle_agent_terminated(*msg[1:])
+                else:
+
+                    pass
+        finally:
+            self._stopped_evt.set()
+
+    async def _init_generation(self) -> None:
+        s = self.state
+        agent_ids = await self._extract_agent_ids(s.population_id)
+        s.agent_ids = agent_ids
+        s.tot_agents = len(agent_ids)
+        s.agents_left = 0
+
+        active = []
+        for aid in agent_ids:
+            try:
+                handle = await self._exoself_start(aid, self)
+                active.append((aid, handle))
+                s.agents_left += 1
+            except Exception as e:
+                log.error(f"[Monitor] FAILED to start agent {aid}: {e!r}")
+
+                await _run("MATCH (a:agent {id:$aid}) SET a.fitness = toFloat($f)", aid=str(aid), f=float("-inf"))
+
+        s.active = active
+        s.tot_agents = s.agents_left
+        log.info(f"*** Population monitor started: pop={s.population_id}, mode={s.op_mode}, "
+                 f"selection={s.selection_algorithm}, agents={s.tot_agents}")
+
+    async def _handle_stop(self, mode: str) -> None:
+        s = self.state
+
+        for (_aid, h) in list(s.active):
+            try:
+                if hasattr(h, "cast"):
+                    await h.cast(("terminate",))
+                elif hasattr(h, "stop"):
+                    await h.stop()
+            except Exception:
+                pass
+
+        for f in (self._episodes_f, self._progress_f):
+            if f is not None:
+                try:
+                    f.flush()
+                    f.close()
+                except Exception:
+                    pass
+        self._episodes_f = None
+        self._progress_f = None
+        log.info(f"*** Population_Monitor:{s.population_id} shutdown. op_tag={s.op_tag}, op_mode={s.op_mode}")
+
+        await neo4j.close()
+
+    async def _handle_episode_done(self, agent_id: str, ep_return: float, eval_idx: int) -> None:
+        s = self.state
+        t_sec = asyncio.get_running_loop().time() - (self._t0 or 0.0)
+        if self._episodes_f is not None:
+            self._episodes_f.write(f"{t_sec:.6f},{s.pop_gen+1},{agent_id},{eval_idx},{ep_return:.10f}\n")
+
+    async def _handle_agent_terminated(self, agent_id: str, fitness: float, agent_eval: int, agent_cycle: int,
+                                       agent_time: int) -> None:
+        log.info(f"agent terminated: , {agent_id}, {fitness}, {agent_eval}, {agent_cycle}, {agent_time}")
+        s = self.state
+
+        s.eval_acc += int(agent_eval)
+        s.cycle_acc += int(agent_cycle)
+        s.time_acc += int(agent_time)
+        s.agents_left -= 1
+
+        await _run("MATCH (a:agent {id:$aid}) SET a.fitness = toFloat($f)", aid=str(agent_id), f=float(fitness))
+
+        s.active = [(aid, h) for (aid, h) in s.active if aid != agent_id]
+
+        log.info(f"[Monitor] agent done: {agent_id} | agents_left={s.agents_left}/{s.tot_agents}")
+        if 0 < s.agents_left <= 3:
+            log.info("[Monitor] still active: %s", [str(aid) for (aid, _h) in s.active])
+
+        if s.agents_left <= 0:
+            await self._generation_finished()
+
+    async def _generation_finished(self) -> None:
+        s = self.state
+        await self._mutate_population(s.population_id, SPECIE_SIZE_LIMIT, s.selection_algorithm)
+        s.pop_gen += 1
+        log.info(f"Population {s.population_id} generation {s.pop_gen} ended.\n")
+
+        pop_stats = await _population_aggregate(s.population_id)
+
+        # aggregate logging information
+        t_sec = asyncio.get_running_loop().time() - (self._t0 or 0.0)
+        t_norm = min(1.0, t_sec / self.train_time_sec)
+        best_gen = float(pop_stats["best"])
+        self._best_so_far = max(self._best_so_far, best_gen)
+
+        # write logs
+        if self._progress_f is not None:
+            self._progress_f.write(
+                f"{t_sec:.6f},{t_norm:.6f},{s.pop_gen},{best_gen:.10f},{self._best_so_far:.10f},"
+                f"{float(pop_stats['avg']):.10f},{float(pop_stats['std']):.10f}\n"
+            )
+
+        s.rows.append({
+            "gen": int(s.pop_gen),
+            "t_sec": int(asyncio.get_running_loop().time()),
+            "cum_fitness": float(pop_stats["cum_fitness"]),
+            "best": float(pop_stats["best"]),
+            "avg": float(pop_stats["avg"]),
+            "std": float(pop_stats["std"]),
+            "agents": int(pop_stats["agents"]),
+            "eval_acc": int(s.eval_acc),
+            "cycle_acc": float(s.cycle_acc),
+            "time_acc": float(s.time_acc),
+        })
+
+        self.gen_ended.set()
+        self.gen_ended = asyncio.Event()
+
+        # check time
+        now = asyncio.get_running_loop().time()
+        elapsed = now - self._t0
+        time_limit_reached = elapsed >= self.train_time_sec
+        if time_limit_reached:
+            log.info(
+                f"[Monitor] time limit reached "
+                f"({elapsed:.1f}s >= {self.train_time_sec}s), stopping run"
+            )
+
+        best = await self._best_fitness_in_population(s.population_id)
+        end_condition = (s.pop_gen >= GENERATION_LIMIT) or (s.eval_acc >= EVALUATIONS_LIMIT) or (best > FITNESS_GOAL) or time_limit_reached
+        if s.pop_gen >= GENERATION_LIMIT:
+            log.info(f"reached generation limit {GENERATION_LIMIT}, stopping")
+        if s.eval_acc >= EVALUATIONS_LIMIT:
+            log.info(f"reached evaluation limit {EVALUATIONS_LIMIT}, stopping")
+        if best > FITNESS_GOAL:
+            log.info(f"reached best fitness {best}, stopping")
+
+        if s.op_tag == "done" or end_condition:
+            await self.inbox.put(("stop", "normal"))
+            return
+
+        if s.op_tag == "pause":
+            log.info("Population Monitor paused.")
+            return
+
+        await self._init_generation()
+
+    async def _ensure_population_node(self, population_id: str) -> None:
+        await _run("MERGE (:population {id:$pid})", pid=str(population_id))
+
+    async def _ensure_specie_node(self, specie_id: str, population_id: str, constraint_json: str) -> None:
+        await _run("""
+            MERGE (s:specie {id:$sid})
+            SET s.population_id = $pid,
+                s.constraint_json = $cjson
+            """, sid=str(specie_id), pid=str(population_id), cjson=str(constraint_json))
+
+    async def _extract_agent_ids(self, population_id: str) -> List[str]:
+        rows = await _read_all("MATCH (a:agent {population_id:$pid}) RETURN a.id AS id ORDER BY id",
+                               pid=str(population_id))
+        return [str(r["id"]) for r in rows]
+
+    async def _extract_specie_ids(self, population_id: str) -> List[str]:
+        rows = await _read_all("""
+            MATCH (s:specie {population_id:$pid}) RETURN s.id AS id ORDER BY id
+            """, pid=str(population_id))
+        return [str(r["id"]) for r in rows]
+
+    async def _extract_specie_agent_ids(self, specie_id: str) -> List[str]:
+        rows = await _read_all("""
+            MATCH (:specie {id:$sid})-[:HAS]->(a:agent) RETURN a.id AS id
+            """, sid=str(specie_id))
+        return [str(r["id"]) for r in rows]
+
+    async def _mutate_population(self, population_id: str, keep_tot: int,
+                                 selection_algorithm: SelectionAlgorithm) -> None:
+        energy_cost = await self._calculate_energy_cost(population_id)
+        specie_ids = await self._extract_specie_ids(population_id)
+        for sid in specie_ids:
+            await self._mutate_specie(sid, keep_tot, energy_cost, selection_algorithm)
+
+    async def _mutate_specie(self, specie_id: str, population_limit: int, neural_energy_cost: float,
+                             selection_algorithm: SelectionAlgorithm) -> None:
+        agent_ids = await self._extract_specie_agent_ids(specie_id)
+
+        summaries = await self._construct_agent_summaries(agent_ids)
+        summaries.sort(key=lambda t: t[0], reverse=True)
+
+        if not summaries:
+            return
+
+        if selection_algorithm == "competition":
+            tot_survivors = round(len(summaries) * SURVIVAL_PERCENTAGE)
+
+            weighted = sorted(
+                [(fit / (max(tn, 1) ** EFF), (fit, tn, aid)) for (fit, tn, aid) in summaries],
+                key=lambda x: x[0], reverse=True
+            )
+            valid = [val for (_score, val) in weighted][:tot_survivors]
+            invalid = [x for x in summaries if x not in valid]
+            top3 = valid[:3]
+            top_agent_ids = [aid for (_f, _tn, aid) in top3]
+
+            for (_f, _tn, aid) in invalid:
+                await delete_agent(aid)
+                await _run("""
+                    MATCH (:specie {id:$sid})-[h:HAS]->(a:agent {id:$aid}) DELETE h
+                """, sid=str(specie_id), aid=str(aid))
+
+            new_ids = await self._competition(valid, population_limit, neural_energy_cost, specie_id)
+
+        elif selection_algorithm == "top3":
+            valid = summaries[:3]
+            invalid = [x for x in summaries if x not in valid]
+            top_agent_ids = [aid for (_f, _tn, aid) in valid]
+            for (_f, _tn, aid) in invalid:
+                await delete_agent(aid)
+                await _run("MATCH (:specie {id:$sid})-[h:HAS]->(a:agent {id:$aid}) DELETE h",
+                           sid=str(specie_id), aid=str(aid))
+
+            offspring_needed = max(0, population_limit - len(top_agent_ids))
+            new_ids = await self._top3(top_agent_ids, offspring_needed, specie_id)
+
+        else:
+            log.error(f"Unknown selection algorithm: {selection_algorithm}")
+            raise ValueError(f"Unknown selection algorithm: {selection_algorithm}")
+
+        f_list = [f for (f, _tn, _aid) in summaries]
+        avg, std, maxf, minf = _mean(f_list), _std(f_list), max(f_list), min(f_list)
+
+        row = await _read_all("MATCH (s:specie {id:$sid}) RETURN toInteger(s.innovation_factor) AS inv",
+                              sid=str(specie_id))
+        inv = int(row[0]["inv"]) if row and row[0]["inv"] is not None else 0
+        u_inv = 0 if (maxf > inv) else (inv - 1)
+
+        await _run("""
+            MATCH (s:specie {id:$sid})
+            SET s.fitness_avg = toFloat($avg),
+                s.fitness_std = toFloat($std),
+                s.fitness_max = toFloat($maxf),
+                s.fitness_min = toFloat($minf),
+                s.innovation_factor = toInteger($inv),
+                s.champion_ids = $champs
+            """, sid=str(specie_id), avg=float(avg), std=float(std), maxf=float(maxf), minf=float(minf),
+                   inv=int(u_inv), champs=[str(x) for x in top_agent_ids])
+
+        for aid in new_ids:
+            try:
+                await update_fingerprint(aid)
+            except Exception:
+                pass
+
+    async def _competition(self, valid: List[Tuple[float, int, str]], population_limit: int,
+                           neural_energy_cost: float, specie_id: str) -> List[str]:
+        alot, est = await self._calculate_alotments(valid, neural_energy_cost)
+        normalizer = (est / population_limit) if population_limit > 0 else 1.0
+        log.debug(f"Population size normalizer: {normalizer:.4f}")
+        return await self._gather_survivors(alot, normalizer, specie_id)
+
+    async def _calculate_alotments(self, valid: List[Tuple[float, int, str]], neural_energy_cost: float
+                                   ) -> Tuple[List[Tuple[float, float, int, str]], float]:
+        acc: List[Tuple[float, float, int, str]] = []
+        new_pop_acc = 0.0
+        for (fit, tn, aid) in valid:
+            neural_alot = (fit / neural_energy_cost) if neural_energy_cost > 0 else 0.0
+            mutant_alot = (neural_alot / max(tn, 1))
+            new_pop_acc += mutant_alot
+            acc.append((mutant_alot, fit, tn, aid))
+        log.debug(f"NewPopAcc: {new_pop_acc:.4f}")
+        return acc, new_pop_acc
+
+    async def _gather_survivors(self, alot: List[Tuple[float, float, int, str]], normalizer: float, specie_id: str) -> \
+            List[str]:
+        new_ids: List[str] = []
+        for (ma, fit, tn, aid) in alot:
+            count = int(round(ma / normalizer)) if normalizer > 0 else 0
+            log.info(f"Agent {aid}: normalized allotment = {count}")
+            if count >= 1:
+
+                await _run("""
+                    MATCH (s:specie {id:$sid}), (a:agent {id:$aid})
+                    MERGE (s)-[:HAS]->(a)
+                """, sid=str(specie_id), aid=str(aid))
+                new_ids.append(aid)
+
+                k = count - 1
+                for _ in range(k):
+                    cid = await self._create_mutant_offspring(aid, specie_id)
+                    new_ids.append(cid)
+            else:
+
+                await delete_agent(aid)
+                await _run("MATCH (:specie {id:$sid})-[h:HAS]->(a:agent {id:$aid}) DELETE h",
+                           sid=str(specie_id), aid=str(aid))
+
+        log.info(f"New Population Size (specie={specie_id}): {len(new_ids)}")
+        return new_ids
+
+    async def _create_mutant_offspring(self, parent_id: str, specie_id: str) -> str:
+        clone_id = _new_id()
+
+        await clone_agent(parent_id, clone_id)
+
+        rows = await _read_all("MATCH (a:agent {id:$aid}) RETURN a.specie_id AS sid, a.population_id AS pid",
+                               aid=str(parent_id))
+        sid = rows[0]["sid"] if rows else specie_id
+        pid = rows[0]["pid"] if rows else None
+        await _run("""
+            MATCH (a:agent {id:$cid})
+            SET a.specie_id = $sid,
+                a.population_id = $pid,
+                a.fitness = NULL
+        """, cid=str(clone_id), sid=str(sid), pid=str(pid))
+
+        await _run("""
+            MATCH (s:specie {id:$sid}), (a:agent {id:$cid})
+            MERGE (s)-[:HAS]->(a)
+        """, sid=str(specie_id), cid=str(clone_id))
+
+        await self._mutate_one_step(clone_id)
+        return clone_id
+
+    async def _mutate_one_step(self, agent_id: str) -> None:
+
+        ops = [
+            self.mutator.mutate_weights,
+            self.mutator.add_bias,
+            self.mutator.remove_bias,
+            self.mutator.add_inlink,
+            self.mutator.add_outlink,
+            self.mutator.add_neuron,
+            self.mutator.outsplice,
+            self.mutator.add_actuator,
+        ]
+        op = random.choice(ops)
+        try:
+            await op(agent_id)
+        except Exception:
+
+            try:
+                await self.mutator.mutate_weights(agent_id)
+            except Exception:
+                pass
+
+    async def _top3(self, valid_ids: List[str], offspring_needed: int, specie_id: str) -> List[str]:
+        new_ids = list(valid_ids)
+        for _ in range(offspring_needed):
+            parent = random.choice(valid_ids)
+            cid = await self._create_mutant_offspring(parent, specie_id)
+            new_ids.append(cid)
+
+        return new_ids
+
+    async def _construct_agent_summaries(self, agent_ids: Sequence[str]) -> List[Tuple[float, int, str]]:
+        if not agent_ids:
+            return []
+        rows = await _read_all("""
+            UNWIND $ids AS aid
+            MATCH (a:agent {id:aid})-[:OWNS]->(cx:cortex)
+            OPTIONAL MATCH (cx)-[:HAS]->(n:neuron)
+            RETURN aid AS id,
+                   toFloat(a.fitness) AS f,
+                   count(n) AS k
+        """, ids=[str(x) for x in agent_ids])
+        out: List[Tuple[float, int, str]] = []
+        for r in rows:
+            f = float(r["f"]) if r["f"] is not None else 0.0
+            k = int(r["k"])
+            out.append((f, k, str(r["id"])))
+        return out
+
+    async def _calculate_energy_cost(self, population_id: str) -> float:
+        rows = await _read_all("""
+                MATCH (a:agent {population_id:$pid})-[:OWNS]->(cx:cortex)
+                OPTIONAL MATCH (cx)-[:HAS]->(n:neuron)
+                RETURN sum(coalesce(toFloat(a.fitness),0.0)) AS totE,
+                       count(n) AS totN
+            """, pid=str(population_id))
+        if not rows:
+            return 0.0
+        totE = float(rows[0]["totE"] or 0.0)
+        totN = int(rows[0]["totN"] or 0)
+        return (totE / totN) if totN > 0 else 0.0
+
+    async def _best_fitness_in_population(self, population_id: str) -> float:
+        rows = await _read_all("""
+            MATCH (a:agent {population_id:$pid})
+            RETURN max(toFloat(a.fitness)) AS best
+        """, pid=str(population_id))
+        return float(rows[0]["best"] or 0.0) if rows else 0.0
+
+
+async def init_population(params: Tuple[str, List[dict], str, SelectionAlgorithm]) -> PopulationMonitor:
+    """
+    params = (Population_Id, Specie_Constraints, OpMode, Selection_Algorithm)
+    - erzeugt Population/Spezies-Knoten
+    - konstruiert Agents (über dein construct_agent)
+    - verknüpft Spezies->Agent, setzt agent.population_id
+    - startet den Monitor
+    """
+    population_id, specie_constraints, op_mode, selection_algorithm = params
+
+    await delete_population(population_id)
+
+    await _run("MERGE (:population {id:$pid})", pid=str(population_id))
+
+    # TODO: this has to move to the genotype module
+    for specon in specie_constraints:
+        sid = _new_id()
+        await _run("""
+            MERGE (p:population {id:$pid})
+            MERGE (s:specie {id:$sid})
+            ON CREATE SET
+                s.population_id   = $pid,
+                s.constraint_json = $cjson,
+                s.innovation_factor = 0
+            ON MATCH SET
+                s.population_id   = $pid,
+                s.constraint_json = $cjson
+            MERGE (p)-[:HAS]->(s)
+        """, pid=str(population_id),
+                   sid=str(sid),
+                   cjson=json.dumps(specon, separators=(",", ":"), sort_keys=True))
+
+        for _ in range(INIT_SPECIE_SIZE):
+            aid = _new_id()
+            await construct_agent(sid, aid, specon)
+
+            # todo: this needs to move to the genotype
+            await _run("""
+                MATCH (a:agent {id:$aid}), (s:specie {id:$sid})
+                SET a.population_id = $pid, a.specie_id = $sid, a.fitness = NULL
+                MERGE (s)-[:HAS]->(a)
+            """, aid=str(aid), sid=str(sid), pid=str(population_id))
+
+    monitor = await PopulationMonitor.start(op_mode, population_id, selection_algorithm)
+    return monitor
+
+
+async def continue_(op_mode: str, selection_algorithm: SelectionAlgorithm,
+                    population_id: str = INIT_POPULATION_ID) -> PopulationMonitor:
+    return await PopulationMonitor.start(op_mode, population_id, selection_algorithm)
+
+
+async def delete_population(population_id: str) -> None:
+    await _run("""
+        MATCH (p:population {id:$pid})
+        OPTIONAL MATCH (s:specie {population_id:$pid})
+        OPTIONAL MATCH (s)-[:HAS]->(a:agent)-[:OWNS]->(cx:cortex)
+        OPTIONAL MATCH (cx)-[:HAS]->(n:neuron)
+        OPTIONAL MATCH (cx)-[:HAS]->(sen:sensor)
+        OPTIONAL MATCH (cx)-[:HAS]->(act:actuator)
+        DETACH DELETE p, s, a, cx, n, sen, act
+    """, pid=str(population_id))
+
+
+async def test() -> PopulationMonitor:
+    return await init_population((INIT_POPULATION_ID, INIT_CONSTRAINTS, INIT_OP_MODE, INIT_SELECTION_ALGO))

mathema/core/trainer.py

@@ -1,11 +1,10 @@
 import asyncio
 import os
-import time
 from typing import Any, Dict, List, Tuple, Optional
 
-import morphology
-from genotype import construct, save_genotype, print_genotype
-from exoself import Exoself
+from mathema.core import morphology
+from mathema.genotype.neo4j.genotype import construct, print_genotype
+from mathema.core.exoself import Exoself
 
 
 class Trainer:

mathema/envs/openai_car_racing.py

@@ -0,0 +1,726 @@
+import math
+import numpy as np
+import logging
+
+import Box2D
+from Box2D import (b2FixtureDef, b2PolygonShape, b2ContactListener)
+
+log = logging.getLogger(__name__)
+
+try:
+    import gymnasium as gym
+    from gymnasium import spaces
+    from gymnasium.utils import seeding
+
+    GYM_AVAILABLE = True
+except Exception:
+    GYM_AVAILABLE = False
+    spaces = None
+    seeding = None
+
+from gymnasium.envs.box2d.car_dynamics import Car
+
+import pygame
+
+DEBUG_DRAWING = False
+LOOK_AHEAD = 10
+
+STATE_W = 96
+STATE_H = 96
+VIDEO_W = 1200
+VIDEO_H = 800
+WINDOW_W = 1350
+WINDOW_H = 950
+
+SCALE = 6.0
+TRACK_RAD = 900 / SCALE
+PLAYFIELD = 2000 / SCALE
+FPS = 60
+ZOOM = 2.7
+ZOOM_FOLLOW = True
+
+TRACK_DETAIL_STEP = 21 / SCALE
+TRACK_TURN_RATE = 0.31
+TRACK_WIDTH = 40 / SCALE
+BORDER = 8 / SCALE
+BORDER_MIN_COUNT = 4
+
+ROAD_COLOR = [0.4, 0.4, 0.4]
+
+MAX_TIME_SEC = 90.0
+MAX_STEPS = int(FPS * MAX_TIME_SEC)
+NO_PROGRESS_SEC = 8.0
+NO_PROGRESS_STEPS = int(FPS * NO_PROGRESS_SEC)
+STALL_MIN_SPEED = 4.0
+STALL_SEC = 4.0
+STALL_STEPS = int(FPS * STALL_SEC)
+FUEL_LIMIT = 120.0
+
+
+def standardize_angle(theta: float) -> float:
+    return np.remainder(theta + np.pi, 2 * np.pi) - np.pi
+
+
+def f2c(rgb_float):
+    """float [0..1] -> int [0..255] color tuple"""
+    return tuple(max(0, min(255, int(255 * x))) for x in rgb_float)
+
+
+class MyState:
+    def __init__(self):
+        self.angle_deltas = None
+        self.reward = None
+        self.on_road = None
+        self.laps = None
+
+        self.wheel_angle = None
+        self.car_angle = None
+        self.angular_vel = None
+        self.true_speed = None
+        self.off_center = None
+        self.vel_angle = None
+
+    def as_array(self, n: int):
+        return np.append(
+            self.angle_deltas[:n],
+            [
+                self.wheel_angle,
+                self.car_angle,
+                self.angular_vel,
+                self.true_speed,
+                self.off_center,
+                self.vel_angle,
+            ],
+        ).astype(np.float32)
+
+    def as_feature_vector(self, lookahead: int = LOOK_AHEAD):
+        return self.as_array(lookahead)
+
+
+class FrictionDetector(b2ContactListener):
+    def __init__(self, env):
+        super().__init__()
+        self.env = env
+
+    def BeginContact(self, contact):
+        self._contact(contact, True)
+
+    def EndContact(self, contact):
+        self._contact(contact, False)
+
+    def _contact(self, contact, begin):
+        tile = None
+        obj = None
+        u1 = contact.fixtureA.body.userData
+        u2 = contact.fixtureB.body.userData
+        if u1 and "road_friction" in u1.__dict__:
+            tile = u1
+            obj = u2
+        if u2 and "road_friction" in u2.__dict__:
+            tile = u2
+            obj = u1
+        if not tile:
+            return
+
+        tile.color[0] = ROAD_COLOR[0]
+        tile.color[1] = ROAD_COLOR[1]
+        tile.color[2] = ROAD_COLOR[2]
+        if not obj or "tiles" not in obj.__dict__:
+            return
+        if begin:
+            obj.tiles.add(tile)
+            if tile.index_on_track == self.env.next_road_tile:
+                self.env.reward += 1000.0 / len(self.env.track)
+                self.env.tile_visited_count += 1
+                self.env.next_road_tile += 1
+                if self.env.next_road_tile >= len(self.env.road):
+                    self.env.next_road_tile = 0
+                    self.env.laps += 1
+        else:
+            if tile in obj.tiles:
+                obj.tiles.remove(tile)
+        self.env.on_road = len(obj.tiles) > 0
+
+
+class CarRacing:
+    metadata = {
+        "render_modes": ["human", "rgb_array", None],
+        "render_fps": FPS,
+    }
+
+    def __init__(self, seed_value: int = 5, render_mode: str | None = "human"):
+
+        self.offroad_frames = None
+        if seeding is not None:
+            self.np_random, _ = seeding.np_random(seed_value)
+        else:
+            self.np_random = np.random.RandomState(seed_value)
+
+        self.contactListener_keepref = FrictionDetector(self)
+        self.world = Box2D.b2World((0, 0), contactListener=self.contactListener_keepref)
+
+        if GYM_AVAILABLE:
+            self.action_space = spaces.Box(
+                np.array([-1, 0, 0], dtype=np.float32),
+                np.array([+1, +1, +1], dtype=np.float32),
+                dtype=np.float32,
+            )
+            self.observation_space = spaces.Box(
+                low=0, high=255, shape=(STATE_H, STATE_W, 3), dtype=np.uint8
+            )
+
+        self.viewer = None
+        self.road = None
+        self.car = None
+        self.reward = 0.0
+        self.prev_reward = 0.0
+
+        self.laps = 0
+        self.on_road = True
+        self.ctrl_pts = None
+        self.outward_vectors = None
+        self.angles = None
+        self.angle_deltas = None
+        self.original_road_poly = None
+        self.indices = None
+        self.my_state = MyState()
+        self.next_road_tile = 0
+
+        self.render_mode = render_mode
+        self._pg = None
+
+        self.tile_visited_count = 0
+        self.t = 0.0
+        self.human_render = False
+
+        self._build_new_episode()
+
+        self.offroad_frames = 0
+        self.offroad_grace_frames = int(0.7 * FPS)
+        self.offroad_penalty_per_frame = 2.0
+
+        self.steps = 0
+        self._last_progress_count = 0
+        self._no_progress_steps = 0
+        self._stall_steps = 0
+
+    class _PygameCtx:
+        def __init__(self):
+            self.initialized = False
+            self.screen = None
+            self.clock = None
+            self.font = None
+            self.rgb_surface = None
+
+    def _ensure_pygame(self):
+        if self._pg is None:
+            self._pg = self._PygameCtx()
+        if not self._pg.initialized:
+            if not pygame.get_init():
+                pygame.init()
+            flags = 0
+            if self.render_mode == "human":
+                self._pg.screen = pygame.display.set_mode((WINDOW_W, WINDOW_H))
+            else:
+
+                self._pg.screen = pygame.Surface((WINDOW_W, WINDOW_H))
+            self._pg.clock = pygame.time.Clock()
+            try:
+                pygame.font.init()
+                self._pg.font = pygame.font.SysFont("Arial", 20)
+            except Exception:
+                self._pg.font = None
+            self._pg.initialized = True
+
+    def _world_to_screen(self, x, y, zoom, angle, scroll_x, scroll_y):
+        ca, sa = math.cos(angle), math.sin(angle)
+
+        rx = (x - scroll_x) * ca + (y - scroll_y) * sa
+        ry = -(x - scroll_x) * sa + (y - scroll_y) * ca
+
+        sx = int(WINDOW_W / 2 + rx * zoom)
+        sy = int(WINDOW_H / 4 + ry * zoom)
+        return sx, sy
+
+    def get_feature_vector(self, lookahead: int = LOOK_AHEAD) -> list[float]:
+        my_s: MyState = self.my_state
+        vec = my_s.as_feature_vector(lookahead).tolist()
+        return vec
+
+    def _draw_polygon_world(self, poly, color, zoom, angle, scroll_x, scroll_y):
+        pts = [self._world_to_screen(px, py, zoom, angle, scroll_x, scroll_y) for (px, py) in poly]
+        pygame.draw.polygon(self._pg.screen, f2c(color), pts)
+
+    def _draw_body(self, body, color=(0.7, 0.7, 0.7), zoom=1.0, angle=0.0, scroll_x=0.0, scroll_y=0.0):
+
+        col = f2c(color)
+        for fixture in body.fixtures:
+            shape = fixture.shape
+            if isinstance(shape, b2PolygonShape):
+                verts = [body.transform * v for v in shape.vertices]
+                pts = [self._world_to_screen(v[0], v[1], zoom, angle, scroll_x, scroll_y) for v in verts]
+                pygame.draw.polygon(self._pg.screen, col, pts, width=0)
+
+    def _destroy(self):
+        if not self.road:
+            return
+
+        for t in self.road:
+            try:
+                t.userData = None
+            except Exception:
+                pass
+            try:
+                self.world.DestroyBody(t)
+            except Exception:
+                pass
+        self.road = []
+
+        if self.car is not None:
+            try:
+                self.car.destroy()
+            except Exception:
+                pass
+            self.car = None
+
+    def _create_track(self):
+        CHECKPOINTS = 12
+        checkpoints = []
+        for c in range(CHECKPOINTS):
+            alpha = 2 * math.pi * c / CHECKPOINTS + self.np_random.uniform(0, 2 * math.pi * 1 / CHECKPOINTS)
+            rad = self.np_random.uniform(TRACK_RAD / 3, TRACK_RAD)
+            if c == 0:
+                alpha = 0
+                rad = 1.5 * TRACK_RAD
+            if c == CHECKPOINTS - 1:
+                alpha = 2 * math.pi * c / CHECKPOINTS
+                self.start_alpha = 2 * math.pi * (-0.5) / CHECKPOINTS
+                rad = 1.5 * TRACK_RAD
+            checkpoints.append((alpha, rad * math.cos(alpha), rad * math.sin(alpha)))
+
+        self.road = []
+
+        x, y, beta = 1.5 * TRACK_RAD, 0, 0
+        dest_i = 0
+        laps = 0
+        track = []
+        no_freeze = 2500
+        visited_other_side = False
+        while True:
+            alpha = math.atan2(y, x)
+            if visited_other_side and alpha > 0:
+                laps += 1
+                visited_other_side = False
+            if alpha < 0:
+                visited_other_side = True
+                alpha += 2 * math.pi
+            while True:
+                failed = True
+                while True:
+                    dest_alpha, dest_x, dest_y = checkpoints[dest_i % len(checkpoints)]
+                    if alpha <= dest_alpha:
+                        failed = False
+                        break
+                    dest_i += 1
+                    if dest_i % len(checkpoints) == 0:
+                        break
+                if not failed:
+                    break
+                alpha -= 2 * math.pi
+                continue
+            r1x = math.cos(beta)
+            r1y = math.sin(beta)
+            p1x = -r1y
+            p1y = r1x
+            dest_dx = dest_x - x
+            dest_dy = dest_y - y
+            proj = r1x * dest_dx + r1y * dest_dy
+            while beta - alpha > 1.5 * math.pi:
+                beta -= 2 * math.pi
+            while beta - alpha < -1.5 * math.pi:
+                beta += 2 * math.pi
+            prev_beta = beta
+            proj *= SCALE
+            if proj > 0.3:
+                beta -= min(TRACK_TURN_RATE, abs(0.001 * proj))
+            if proj < -0.3:
+                beta += min(TRACK_TURN_RATE, abs(0.001 * proj))
+            x += p1x * TRACK_DETAIL_STEP
+            y += p1y * TRACK_DETAIL_STEP
+            track.append((alpha, prev_beta * 0.5 + beta * 0.5, x, y))
+            if laps > 4:
+                break
+            no_freeze -= 1
+            if no_freeze == 0:
+                break
+
+        i1, i2 = -1, -1
+        i = len(track)
+        while True:
+            i -= 1
+            if i == 0:
+                return False
+            pass_through_start = track[i][0] > self.start_alpha >= track[i - 1][0]
+            if pass_through_start and i2 == -1:
+                i2 = i
+            elif pass_through_start and i1 == -1:
+                i1 = i
+                break
+        print(f"Track generation: {i1}..{i2} -> {i2 - i1}-tiles track")
+        assert i1 != -1 and i2 != -1
+
+        track = track[i1: i2 - 1]
+
+        first_beta = track[0][1]
+        first_perp_x = math.cos(first_beta)
+        first_perp_y = math.sin(first_beta)
+        well_glued_together = np.sqrt(
+            np.square(first_perp_x * (track[0][2] - track[-1][2]))
+            + np.square(first_perp_y * (track[0][3] - track[-1][3]))
+        )
+        if well_glued_together > TRACK_DETAIL_STEP:
+            return False
+
+        border = [False] * len(track)
+        for i in range(len(track)):
+            good = True
+            oneside = 0
+            for neg in range(BORDER_MIN_COUNT):
+                beta1 = track[i - neg - 0][1]
+                beta2 = track[i - neg - 1][1]
+                good &= abs(beta1 - beta2) > TRACK_TURN_RATE * 0.2
+                oneside += np.sign(beta1 - beta2)
+            good &= abs(oneside) == BORDER_MIN_COUNT
+            border[i] = bool(good)
+        for i in range(len(track)):
+            for neg in range(BORDER_MIN_COUNT):
+                border[i - neg] |= border[i]
+
+        self.road_poly = []
+        for i in range(len(track)):
+            alpha1, beta1, x1, y1 = track[i]
+            alpha2, beta2, x2, y2 = track[i - 1]
+            road1_l = (x1 - TRACK_WIDTH * math.cos(beta1), y1 - TRACK_WIDTH * math.sin(beta1))
+            road1_r = (x1 + TRACK_WIDTH * math.cos(beta1), y1 + TRACK_WIDTH * math.sin(beta1))
+            road2_l = (x2 - TRACK_WIDTH * math.cos(beta2), y2 - TRACK_WIDTH * math.sin(beta2))
+            road2_r = (x2 + TRACK_WIDTH * math.cos(beta2), y2 + TRACK_WIDTH * math.sin(beta2))
+            t = self.world.CreateStaticBody(
+                fixtures=b2FixtureDef(shape=b2PolygonShape(vertices=[road1_l, road1_r, road2_r, road2_l]))
+            )
+            t.userData = t
+            t.index_on_track = i
+            c = 0.01 * (i % 3)
+            t.color = [ROAD_COLOR[0] + c, ROAD_COLOR[1] + c, ROAD_COLOR[2] + c]
+            t.road_visited = False
+            t.road_friction = 1.0
+            t.fixtures[0].sensor = True
+            self.road_poly.append(([road1_l, road1_r, road2_r, road2_l], t.color))
+            self.road.append(t)
+            if border[i]:
+                side = np.sign(beta2 - beta1)
+                b1_l = (x1 + side * TRACK_WIDTH * math.cos(beta1), y1 + side * TRACK_WIDTH * math.sin(beta1))
+                b1_r = (
+                    x1 + side * (TRACK_WIDTH + BORDER) * math.cos(beta1),
+                    y1 + side * (TRACK_WIDTH + BORDER) * math.sin(beta1),
+                )
+                b2_l = (x2 + side * TRACK_WIDTH * math.cos(beta2), y2 + side * TRACK_WIDTH * math.sin(beta2))
+                b2_r = (
+                    x2 + side * (TRACK_WIDTH + BORDER) * math.cos(beta2),
+                    y2 + side * (TRACK_WIDTH + BORDER) * math.sin(beta2),
+                )
+                self.road_poly.append(([b1_l, b1_r, b2_r, b2_l], (1, 1, 1) if i % 2 == 0 else (1, 0, 0)))
+        self.track = track
+
+        self.original_road_poly = [((list(poly)), list(color)) for (poly, color) in self.road_poly]
+        self.ctrl_pts = np.array(list(map(lambda x: x[2:], self.track)))
+        self.angles = np.array(list(map(lambda x: x[1], self.track)))
+        self.outward_vectors = [np.array([np.cos(theta), np.sin(theta)]) for theta in self.angles]
+        angle_deltas = self.angles - np.roll(self.angles, 1)
+        self.angle_deltas = np.array(list(map(standardize_angle, angle_deltas)))
+        self.indices = np.array(range(len(self.ctrl_pts)))
+        return True
+
+    def _build_new_episode(self):
+
+        self._destroy()
+        self.reward = 0.0
+        self.prev_reward = 0.0
+        self.tile_visited_count = 0
+        self.t = 0.0
+        self.road_poly = []
+        self.human_render = False
+        self.laps = 0
+        self.on_road = True
+        self.next_road_tile = 0
+
+        while True:
+            success = self._create_track()
+            if success:
+                break
+            print("retry to generate track (normal if there are not many of this messages)")
+
+        self.car = Car(self.world, *self.track[0][1:4])
+
+        self.car.tiles = set()
+
+        self.steps = 0
+        self._last_progress_count = 0
+        self._no_progress_steps = 0
+        self._stall_steps = 0
+
+    def reset(self, *, seed: int | None = None, options: dict | None = None):
+        if seed is not None:
+
+            if seeding is not None:
+                self.np_random, _ = seeding.np_random(seed)
+            else:
+                self.np_random = np.random.RandomState(seed)
+        self._build_new_episode()
+        obs = self._get_observation()
+        info = {}
+        return obs, info
+
+    def fast_reset(self):
+
+        self.car = None
+        self.laps = 0
+        self.on_road = True
+        self.next_road_tile = 0
+
+        self.reward = 0.0
+        self.prev_reward = 0.0
+        self.tile_visited_count = 0
+        self.t = 0.0
+        self.human_render = False
+        for tile in self.road:
+            tile.road_visited = False
+        self.road_poly = [((list(poly)), list(color)) for (poly, color) in self.original_road_poly]
+        try:
+            self.car.destroy()
+        except Exception:
+            pass
+        self.car = Car(self.world, *self.track[0][1:4])
+        self.car.tiles = set()
+
+        self.steps = 0
+        self._last_progress_count = 0
+        self._no_progress_steps = 0
+        self._stall_steps = 0
+
+        return self.step(np.array([0.0, 0.0, 0.0], dtype=np.float32))
+
+    def step(self, action):
+
+        if action is not None:
+            self.car.steer(-float(action[0]))
+            self.car.gas(float(action[1]))
+            self.car.brake(float(action[2]))
+
+        self.car.step(1.0 / FPS)
+        self.world.Step(1.0 / FPS, 6 * 30, 2 * 30)
+        self.t += 1.0 / FPS
+
+        self.steps += 1
+
+        terminated = False
+        truncated = False
+
+        if action is not None:
+
+            self.reward -= 5.0 / FPS
+
+            if self.tile_visited_count == len(self.track):
+                terminated = True
+
+            x, y = self.car.hull.position
+            if abs(x) > PLAYFIELD or abs(y) > PLAYFIELD:
+                self.reward -= 100.0
+                terminated = True
+
+            if not self.on_road:
+                self.offroad_frames += 1
+                self.reward -= self.offroad_penalty_per_frame / FPS
+                if self.offroad_frames > self.offroad_grace_frames:
+                    self.reward -= 20.0
+                    terminated = True
+            else:
+                self.offroad_frames = 0
+
+            if self.tile_visited_count > self._last_progress_count:
+                self._last_progress_count = self.tile_visited_count
+                self._no_progress_steps = 0
+            else:
+                self._no_progress_steps += 1
+            if self._no_progress_steps >= NO_PROGRESS_STEPS:
+                truncated = True
+
+            step_reward = self.reward - self.prev_reward
+            self.prev_reward = self.reward
+        else:
+            step_reward = 0.0
+
+        v1 = self.outward_vectors[self.next_road_tile - 2]
+        v2 = np.array(self.car.hull.position) - self.ctrl_pts[self.next_road_tile - 1]
+        off_center = float(np.dot(v1, v2))
+        angular_vel = float(self.car.hull.angularVelocity)
+        vel = self.car.hull.linearVelocity
+        true_speed = float(np.linalg.norm(vel))
+        car_angle = float(self.car.hull.angle - self.angles[self.next_road_tile])
+        wheel_angle = float(self.car.wheels[0].joint.angle)
+        if true_speed < 0.2:
+            vel_angle = 0.0
+        else:
+            vel_angle = float(math.atan2(vel[1], vel[0]) - (self.angles[self.next_road_tile] + np.pi / 2))
+
+        wheel_angle = standardize_angle(wheel_angle)
+        car_angle = standardize_angle(car_angle)
+        vel_angle = standardize_angle(vel_angle)
+
+        tip = np.array((self.car.wheels[0].position + self.car.wheels[1].position) / 2)
+        p1 = self.ctrl_pts[self.next_road_tile - 1]
+        p2 = self.ctrl_pts[self.next_road_tile - 2]
+        u = (p1 - p2) / TRACK_DETAIL_STEP
+        v = (tip - p2) / TRACK_DETAIL_STEP
+        interp = float(np.dot(v, u))
+        interp_angle_deltas = np.interp(self.indices + interp, self.indices, self.angle_deltas)
+
+        self.my_state.angle_deltas = np.roll(interp_angle_deltas, -self.next_road_tile)
+        self.my_state.reward = self.reward
+        self.my_state.on_road = self.on_road
+        self.my_state.laps = self.laps
+        self.my_state.true_speed = true_speed
+        self.my_state.off_center = off_center
+        self.my_state.wheel_angle = wheel_angle
+        self.my_state.car_angle = car_angle
+        self.my_state.angular_vel = angular_vel
+        self.my_state.vel_angle = vel_angle
+
+        self.my_state.angle_deltas *= 2.3
+        self.my_state.true_speed /= 100.0
+        self.my_state.off_center /= TRACK_WIDTH
+        self.my_state.wheel_angle *= 2.1
+        self.my_state.car_angle *= 1.5
+        self.my_state.vel_angle *= 1.5
+        self.my_state.angular_vel /= 3.74
+
+        obs = self._get_observation()
+        info = {"features": self.my_state}
+        return obs, step_reward, terminated, truncated, info
+
+    def _get_observation(self):
+
+        return None
+
+    def render(self):
+        self._ensure_pygame()
+
+        if self.render_mode == "human":
+            for event in pygame.event.get():
+                if event.type == pygame.QUIT:
+                    self.close()
+                    return None
+
+        zoom = 0.1 * SCALE * max(1 - self.t, 0) + ZOOM * SCALE * min(self.t, 1)
+        scroll_x = self.car.hull.position[0]
+        scroll_y = self.car.hull.position[1]
+        angle = -self.car.hull.angle
+        vel = self.car.hull.linearVelocity
+        if np.linalg.norm(vel) > 0.5:
+            angle = math.atan2(vel[0], vel[1])
+
+        self._pg.screen.fill((102, 230, 102))
+
+        k = PLAYFIELD / 20.0
+        grid_color = (110, 240, 110)
+        for x in range(-20, 20, 2):
+            for y in range(-20, 20, 2):
+                x0, y0 = k * x + 0, k * y + 0
+                x1, y1 = k * x + k, k * y + k
+                p0 = self._world_to_screen(x0, y0, zoom, angle, scroll_x, scroll_y)
+                p1 = self._world_to_screen(x1, y0, zoom, angle, scroll_x, scroll_y)
+                p2 = self._world_to_screen(x1, y1, zoom, angle, scroll_x, scroll_y)
+                p3 = self._world_to_screen(x0, y1, zoom, angle, scroll_x, scroll_y)
+                pygame.draw.polygon(self._pg.screen, grid_color, [p0, p1, p2, p3])
+
+        for poly, color in self.road_poly:
+            self._draw_polygon_world(poly, color, zoom, angle, scroll_x, scroll_y)
+
+        car_col = (0.25, 0.25, 0.25)
+        self._draw_body(self.car.hull, car_col, zoom, angle, scroll_x, scroll_y)
+        for w in self.car.wheels:
+            self._draw_body(w, (0.15, 0.15, 0.15), zoom, angle, scroll_x, scroll_y)
+
+        if self._pg.font is not None:
+            txt = f"reward={self.reward:0.1f}  laps={self.laps}"
+            surf = self._pg.font.render(txt, True, (255, 255, 255))
+            self._pg.screen.blit(surf, (10, 10))
+
+        if self.render_mode == "human":
+            pygame.display.flip()
+            self._pg.clock.tick(FPS)
+            return None
+        else:
+
+            arr = pygame.surfarray.array3d(self._pg.screen)
+            arr = np.transpose(arr, (1, 0, 2))
+            return arr
+
+    def close(self):
+        try:
+            if self._pg and self._pg.initialized:
+                pygame.display.quit()
+                pygame.quit()
+        except Exception:
+            pass
+        self._pg = None
+
+
+if __name__ == "__main__":
+    import pygame
+
+    pygame.init()
+    env = CarRacing(render_mode="human")
+
+    action = np.array([0.0, 0.0, 0.0], dtype=np.float32)
+    running = True
+
+
+    def handle_keys(a):
+        keys = pygame.key.get_pressed()
+        steer = 0.0
+        if keys[pygame.K_LEFT]:
+            steer -= 1.0
+        if keys[pygame.K_RIGHT]:
+            steer += 1.0
+        gas = 1.0 if keys[pygame.K_UP] else 0.0
+        brake = 0.5 if keys[pygame.K_DOWN] else 0.0
+        a[0], a[1], a[2] = steer, gas, brake
+
+
+    env.reset()
+    try:
+        while running:
+            for event in pygame.event.get():
+                if event.type == pygame.QUIT:
+                    running = False
+                if event.type == pygame.KEYDOWN and event.key == pygame.K_RETURN:
+                    env.fast_reset()
+
+            handle_keys(action)
+            obs, r, terminated, truncated, info = env.step(action)
+
+            if int(env.t * FPS) % 200 == 0:
+                ms: MyState = info.get("features")
+                if ms is not None:
+                    print(
+                        f"speed={ms.true_speed:5.2f} off_center={ms.off_center:+.2f} car_ang={ms.car_angle:+.2f} "
+                        f"reward={r:+.2f}"
+                        f"reward={ms.reward:+.2f}"
+                    )
+
+            env.render()
+            if terminated or truncated:
+                env.fast_reset()
+
+    finally:
+        env.close()

mathema/envs/openai_car_racing_sac.py

@@ -0,0 +1,790 @@
+# TODO: this will be one env for both systems
+
+import math
+import numpy as np
+import logging
+
+import Box2D
+from Box2D import (b2FixtureDef, b2PolygonShape, b2ContactListener)
+
+log = logging.getLogger(__name__)
+
+# Optional gym import for spaces only; code runs without strict gym registration
+try:
+    import gymnasium as gym
+    from gymnasium import spaces
+    from gymnasium.utils import seeding
+
+    GYM_AVAILABLE = True
+except Exception:
+    GYM_AVAILABLE = False
+    spaces = None
+    seeding = None
+
+# Car dynamics from classic gym (Box2D)
+from gymnasium.envs.box2d.car_dynamics import Car
+
+# --- pygame renderer ---
+import pygame
+
+DEBUG_DRAWING = False
+LOOK_AHEAD = 10
+
+STATE_W = 96
+STATE_H = 96
+VIDEO_W = 1200
+VIDEO_H = 800
+WINDOW_W = 1350
+WINDOW_H = 950
+
+SCALE = 6.0
+TRACK_RAD = 900 / SCALE
+PLAYFIELD = 2000 / SCALE
+FPS = 60
+ZOOM = 2.7
+ZOOM_FOLLOW = True
+
+TRACK_DETAIL_STEP = 21 / SCALE
+TRACK_TURN_RATE = 0.31
+TRACK_WIDTH = 40 / SCALE
+BORDER = 8 / SCALE
+BORDER_MIN_COUNT = 4
+
+ROAD_COLOR = [0.4, 0.4, 0.4]
+
+# limits & timeouts
+MAX_TIME_SEC = 90.0
+MAX_STEPS = int(FPS * MAX_TIME_SEC)
+NO_PROGRESS_SEC = 8.0
+NO_PROGRESS_STEPS = int(FPS * NO_PROGRESS_SEC)
+STALL_MIN_SPEED = 4.0
+STALL_SEC = 4.0
+STALL_STEPS = int(FPS * STALL_SEC)
+FUEL_LIMIT = 120.0
+
+
+# ----------------------------
+# Utilities
+# ----------------------------
+def standardize_angle(theta: float) -> float:
+    return np.remainder(theta + np.pi, 2 * np.pi) - np.pi
+
+
+def f2c(rgb_float):
+    """float [0..1] -> int [0..255] color tuple"""
+    return tuple(max(0, min(255, int(255 * x))) for x in rgb_float)
+
+
+# ----------------------------
+# MyState: feature container
+# ----------------------------
+class MyState:
+    def __init__(self):
+        self.angle_deltas = None
+        self.reward = None
+        self.on_road = None
+        self.laps = None
+
+        self.wheel_angle = None
+        self.car_angle = None
+        self.angular_vel = None
+        self.true_speed = None
+        self.off_center = None
+        self.vel_angle = None
+
+    def as_array(self, n: int):
+        return np.append(
+            self.angle_deltas[:n],
+            [
+                self.wheel_angle,
+                self.car_angle,
+                self.angular_vel,
+                self.true_speed,
+                self.off_center,
+                self.vel_angle,
+            ],
+        ).astype(np.float32)
+
+    def as_feature_vector(self, lookahead: int = LOOK_AHEAD):
+        return self.as_array(lookahead)
+
+
+# ----------------------------
+# Contact listener: counts tiles, progress & reward
+# ----------------------------
+class FrictionDetector(b2ContactListener):
+    def __init__(self, env):
+        super().__init__()
+        self.env = env
+
+    def BeginContact(self, contact):
+        self._contact(contact, True)
+
+    def EndContact(self, contact):
+        self._contact(contact, False)
+
+    def _contact(self, contact, begin):
+        tile = None
+        obj = None
+        u1 = contact.fixtureA.body.userData
+        u2 = contact.fixtureB.body.userData
+        if u1 and "road_friction" in u1.__dict__:
+            tile = u1
+            obj = u2
+        if u2 and "road_friction" in u2.__dict__:
+            tile = u2
+            obj = u1
+        if not tile:
+            return
+
+        tile.color[0] = ROAD_COLOR[0]
+        tile.color[1] = ROAD_COLOR[1]
+        tile.color[2] = ROAD_COLOR[2]
+        if not obj or "tiles" not in obj.__dict__:
+            return
+        if begin:
+            obj.tiles.add(tile)
+            if tile.index_on_track == self.env.next_road_tile:
+                self.env.reward += 1000.0 / len(self.env.track)
+                self.env.tile_visited_count += 1
+                self.env.next_road_tile += 1
+                if self.env.next_road_tile >= len(self.env.road):
+                    self.env.next_road_tile = 0
+                    self.env.laps += 1
+        else:
+            if tile in obj.tiles:
+                obj.tiles.remove(tile)
+        self.env.on_road = len(obj.tiles) > 0
+
+
+# ----------------------------
+# CarRacing with pygame rendering and Gym(nasium) 0.26+ compatible step/reset
+# ----------------------------
+class CarRacing(gym.Env):
+    metadata = {
+        "render_modes": ["human", "rgb_array", None],
+        "render_fps": FPS,
+    }
+
+    def __init__(self, seed_value: int = 5, render_mode: str | None = "human"):
+        # RNG
+        self.offroad_frames = None
+        if seeding is not None:
+            self.np_random, _ = seeding.np_random(seed_value)
+        else:
+            self.np_random = np.random.RandomState(seed_value)
+
+        # Physics world
+        self.contactListener_keepref = FrictionDetector(self)
+        self.world = Box2D.b2World((0, 0), contactListener=self.contactListener_keepref)
+
+        # Gym-style spaces (optional)
+        if GYM_AVAILABLE:
+            self.action_space = spaces.Box(
+                np.array([-1, 0, 0], dtype=np.float32),
+                np.array([+1, +1, +1], dtype=np.float32),
+                dtype=np.float32,
+            )
+            # Feature-Vektor-Länge = LOOK_AHEAD + 6 (wheel, car, angular_vel, true_speed, off_center, vel_angle)
+            feat_len = LOOK_AHEAD + 6
+            self.observation_space = spaces.Box(
+                low=-np.inf, high=np.inf, shape=(feat_len,), dtype=np.float32
+            )
+
+        # State
+        self.viewer = None  # unused (pyglet placeholder)
+        self.road = None
+        self.car = None
+        self.reward = 0.0
+        self.prev_reward = 0.0
+
+        self.laps = 0
+        self.on_road = True
+        self.ctrl_pts = None
+        self.outward_vectors = None
+        self.angles = None
+        self.angle_deltas = None
+        self.original_road_poly = None
+        self.indices = None
+        self.my_state = MyState()
+        self.next_road_tile = 0
+
+        # Rendering
+        self.render_mode = render_mode
+        self._pg = None  # pygame objects container
+
+        # Episode control
+        self.tile_visited_count = 0
+        self.t = 0.0
+        self.human_render = False
+
+        # Build initial track + car
+        self._build_new_episode()
+
+        self.offroad_frames = 0
+        self.offroad_grace_frames = int(0.2 * FPS)
+        self.offroad_penalty_per_frame = 2.0
+
+        self.steps = 0
+        self._last_progress_count = 0
+        self._no_progress_steps = 0
+        self._stall_steps = 0
+
+    # ------------------------
+    # Helpers: pygame
+    # ------------------------
+    class _PygameCtx:
+        def __init__(self):
+            self.initialized = False
+            self.screen = None
+            self.clock = None
+            self.font = None
+            self.rgb_surface = None  # offscreen for rgb_array
+
+    def _ensure_pygame(self):
+        if self._pg is None:
+            self._pg = self._PygameCtx()
+        if not self._pg.initialized:
+            if not pygame.get_init():
+                pygame.init()
+            flags = 0
+            if self.render_mode == "human":
+                self._pg.screen = pygame.display.set_mode((WINDOW_W, WINDOW_H))
+            else:
+                # offscreen surface; we can still blit/draw onto it
+                self._pg.screen = pygame.Surface((WINDOW_W, WINDOW_H))
+            self._pg.clock = pygame.time.Clock()
+            try:
+                pygame.font.init()
+                self._pg.font = pygame.font.SysFont("Arial", 20)
+            except Exception:
+                self._pg.font = None
+            self._pg.initialized = True
+
+    def _world_to_screen(self, x, y, zoom, angle, scroll_x, scroll_y):
+        ca, sa = math.cos(angle), math.sin(angle)
+        # rotate around (scroll_x, scroll_y)
+        rx = (x - scroll_x) * ca + (y - scroll_y) * sa
+        ry = -(x - scroll_x) * sa + (y - scroll_y) * ca
+        # scale & translate (match original camera placement)
+        sx = int(WINDOW_W / 2 + rx * zoom)
+        sy = int(WINDOW_H / 4 + ry * zoom)
+        return sx, sy
+
+    def get_feature_vector(self, lookahead: int = LOOK_AHEAD) -> list[float]:
+        my_s: MyState = self.my_state
+        vec = my_s.as_feature_vector(lookahead).tolist()
+        return vec
+
+    def _draw_polygon_world(self, poly, color, zoom, angle, scroll_x, scroll_y):
+        pts = [self._world_to_screen(px, py, zoom, angle, scroll_x, scroll_y) for (px, py) in poly]
+        pygame.draw.polygon(self._pg.screen, f2c(color), pts)
+
+    def _draw_body(self, body, color=(0.7, 0.7, 0.7), zoom=1.0, angle=0.0, scroll_x=0.0, scroll_y=0.0):
+        # Draw each fixture polygon
+        col = f2c(color)
+        for fixture in body.fixtures:
+            shape = fixture.shape
+            if isinstance(shape, b2PolygonShape):
+                verts = [body.transform * v for v in shape.vertices]
+                pts = [self._world_to_screen(v[0], v[1], zoom, angle, scroll_x, scroll_y) for v in verts]
+                pygame.draw.polygon(self._pg.screen, col, pts, width=0)
+
+    # ------------------------
+    # Track & episode setup
+    # ------------------------
+    def _destroy(self):
+        if not self.road:
+            return
+        # userData lösen, dann Bodies zerstören
+        for t in self.road:
+            try:
+                t.userData = None
+            except Exception:
+                pass
+            try:
+                self.world.DestroyBody(t)
+            except Exception:
+                pass
+        self.road = []
+
+        if self.car is not None:
+            try:
+                self.car.destroy()
+            except Exception:
+                pass
+            self.car = None
+
+    def _create_track(self):
+        CHECKPOINTS = 12
+        checkpoints = []
+        for c in range(CHECKPOINTS):
+            alpha = 2 * math.pi * c / CHECKPOINTS + self.np_random.uniform(0, 2 * math.pi * 1 / CHECKPOINTS)
+            rad = self.np_random.uniform(TRACK_RAD / 3, TRACK_RAD)
+            if c == 0:
+                alpha = 0
+                rad = 1.5 * TRACK_RAD
+            if c == CHECKPOINTS - 1:
+                alpha = 2 * math.pi * c / CHECKPOINTS
+                self.start_alpha = 2 * math.pi * (-0.5) / CHECKPOINTS
+                rad = 1.5 * TRACK_RAD
+            checkpoints.append((alpha, rad * math.cos(alpha), rad * math.sin(alpha)))
+
+        self.road = []
+
+        x, y, beta = 1.5 * TRACK_RAD, 0, 0
+        dest_i = 0
+        laps = 0
+        track = []
+        no_freeze = 2500
+        visited_other_side = False
+        while True:
+            alpha = math.atan2(y, x)
+            if visited_other_side and alpha > 0:
+                laps += 1
+                visited_other_side = False
+            if alpha < 0:
+                visited_other_side = True
+                alpha += 2 * math.pi
+            while True:
+                failed = True
+                while True:
+                    dest_alpha, dest_x, dest_y = checkpoints[dest_i % len(checkpoints)]
+                    if alpha <= dest_alpha:
+                        failed = False
+                        break
+                    dest_i += 1
+                    if dest_i % len(checkpoints) == 0:
+                        break
+                if not failed:
+                    break
+                alpha -= 2 * math.pi
+                continue
+            r1x = math.cos(beta)
+            r1y = math.sin(beta)
+            p1x = -r1y
+            p1y = r1x
+            dest_dx = dest_x - x
+            dest_dy = dest_y - y
+            proj = r1x * dest_dx + r1y * dest_dy
+            while beta - alpha > 1.5 * math.pi:
+                beta -= 2 * math.pi
+            while beta - alpha < -1.5 * math.pi:
+                beta += 2 * math.pi
+            prev_beta = beta
+            proj *= SCALE
+            if proj > 0.3:
+                beta -= min(TRACK_TURN_RATE, abs(0.001 * proj))
+            if proj < -0.3:
+                beta += min(TRACK_TURN_RATE, abs(0.001 * proj))
+            x += p1x * TRACK_DETAIL_STEP
+            y += p1y * TRACK_DETAIL_STEP
+            track.append((alpha, prev_beta * 0.5 + beta * 0.5, x, y))
+            if laps > 4:
+                break
+            no_freeze -= 1
+            if no_freeze == 0:
+                break
+
+        i1, i2 = -1, -1
+        i = len(track)
+        while True:
+            i -= 1
+            if i == 0:
+                return False
+            pass_through_start = track[i][0] > self.start_alpha >= track[i - 1][0]
+            if pass_through_start and i2 == -1:
+                i2 = i
+            elif pass_through_start and i1 == -1:
+                i1 = i
+                break
+        print(f"Track generation: {i1}..{i2} -> {i2 - i1}-tiles track")
+        assert i1 != -1 and i2 != -1
+
+        track = track[i1: i2 - 1]
+
+        first_beta = track[0][1]
+        first_perp_x = math.cos(first_beta)
+        first_perp_y = math.sin(first_beta)
+        well_glued_together = np.sqrt(
+            np.square(first_perp_x * (track[0][2] - track[-1][2]))
+            + np.square(first_perp_y * (track[0][3] - track[-1][3]))
+        )
+        if well_glued_together > TRACK_DETAIL_STEP:
+            return False
+
+        border = [False] * len(track)
+        for i in range(len(track)):
+            good = True
+            oneside = 0
+            for neg in range(BORDER_MIN_COUNT):
+                beta1 = track[i - neg - 0][1]
+                beta2 = track[i - neg - 1][1]
+                good &= abs(beta1 - beta2) > TRACK_TURN_RATE * 0.2
+                oneside += np.sign(beta1 - beta2)
+            good &= abs(oneside) == BORDER_MIN_COUNT
+            border[i] = bool(good)
+        for i in range(len(track)):
+            for neg in range(BORDER_MIN_COUNT):
+                border[i - neg] |= border[i]
+
+        self.road_poly = []
+        for i in range(len(track)):
+            alpha1, beta1, x1, y1 = track[i]
+            alpha2, beta2, x2, y2 = track[i - 1]
+            road1_l = (x1 - TRACK_WIDTH * math.cos(beta1), y1 - TRACK_WIDTH * math.sin(beta1))
+            road1_r = (x1 + TRACK_WIDTH * math.cos(beta1), y1 + TRACK_WIDTH * math.sin(beta1))
+            road2_l = (x2 - TRACK_WIDTH * math.cos(beta2), y2 - TRACK_WIDTH * math.sin(beta2))
+            road2_r = (x2 + TRACK_WIDTH * math.cos(beta2), y2 + TRACK_WIDTH * math.sin(beta2))
+            t = self.world.CreateStaticBody(
+                fixtures=b2FixtureDef(shape=b2PolygonShape(vertices=[road1_l, road1_r, road2_r, road2_l]))
+            )
+            t.userData = t
+            t.index_on_track = i
+            c = 0.01 * (i % 3)
+            t.color = [ROAD_COLOR[0] + c, ROAD_COLOR[1] + c, ROAD_COLOR[2] + c]
+            t.road_visited = False
+            t.road_friction = 1.0
+            t.fixtures[0].sensor = True
+            self.road_poly.append(([road1_l, road1_r, road2_r, road2_l], t.color))
+            self.road.append(t)
+            if border[i]:
+                side = np.sign(beta2 - beta1)
+                b1_l = (x1 + side * TRACK_WIDTH * math.cos(beta1), y1 + side * TRACK_WIDTH * math.sin(beta1))
+                b1_r = (
+                    x1 + side * (TRACK_WIDTH + BORDER) * math.cos(beta1),
+                    y1 + side * (TRACK_WIDTH + BORDER) * math.sin(beta1),
+                )
+                b2_l = (x2 + side * TRACK_WIDTH * math.cos(beta2), y2 + side * TRACK_WIDTH * math.sin(beta2))
+                b2_r = (
+                    x2 + side * (TRACK_WIDTH + BORDER) * math.cos(beta2),
+                    y2 + side * (TRACK_WIDTH + BORDER) * math.sin(beta2),
+                )
+                self.road_poly.append(([b1_l, b1_r, b2_r, b2_l], (1, 1, 1) if i % 2 == 0 else (1, 0, 0)))
+        self.track = track
+
+        self.original_road_poly = [((list(poly)), list(color)) for (poly, color) in self.road_poly]
+        self.ctrl_pts = np.array(list(map(lambda x: x[2:], self.track)))
+        self.angles = np.array(list(map(lambda x: x[1], self.track)))
+        self.outward_vectors = [np.array([np.cos(theta), np.sin(theta)]) for theta in self.angles]
+        angle_deltas = self.angles - np.roll(self.angles, 1)
+        self.angle_deltas = np.array(list(map(standardize_angle, angle_deltas)))
+        self.indices = np.array(range(len(self.ctrl_pts)))
+        return True
+
+    def _build_new_episode(self):
+        # build track (may retry)
+        self._destroy()
+        self.reward = 0.0
+        self.prev_reward = 0.0
+        self.tile_visited_count = 0
+        self.t = 0.0
+        self.road_poly = []
+        self.human_render = False
+        self.laps = 0
+        self.on_road = True
+        self.next_road_tile = 0
+
+        while True:
+            success = self._create_track()
+            if success:
+                break
+            print("retry to generate track (normal if there are not many of this messages)")
+
+        self.car = Car(self.world, *self.track[0][1:4])
+
+        # attach tiles set to car for contact tracking
+        self.car.tiles = set()
+
+        self.steps = 0
+        self._last_progress_count = 0
+        self._no_progress_steps = 0
+        self._stall_steps = 0
+
+    # ------------------------
+    # Public API (Gym 0.26+/Gymnasium style)
+    # ------------------------
+    def _update_features(self):
+        v1 = self.outward_vectors[self.next_road_tile - 2]
+        v2 = np.array(self.car.hull.position) - self.ctrl_pts[self.next_road_tile - 1]
+        off_center = float(np.dot(v1, v2))
+        angular_vel = float(self.car.hull.angularVelocity)
+        vel = self.car.hull.linearVelocity
+        true_speed = float(np.linalg.norm(vel))
+        car_angle = float(self.car.hull.angle - self.angles[self.next_road_tile])
+        wheel_angle = float(self.car.wheels[0].joint.angle)
+        if true_speed < 0.2:
+            vel_angle = 0.0
+        else:
+            vel_angle = float(math.atan2(vel[1], vel[0]) - (self.angles[self.next_road_tile] + np.pi / 2))
+
+        wheel_angle = standardize_angle(wheel_angle)
+        car_angle = standardize_angle(car_angle)
+        vel_angle = standardize_angle(vel_angle)
+
+        tip = np.array((self.car.wheels[0].position + self.car.wheels[1].position) / 2)
+        p1 = self.ctrl_pts[self.next_road_tile - 1]
+        p2 = self.ctrl_pts[self.next_road_tile - 2]
+        u = (p1 - p2) / TRACK_DETAIL_STEP
+        v = (tip - p2) / TRACK_DETAIL_STEP
+        interp = float(np.dot(v, u))
+        interp_angle_deltas = np.interp(self.indices + interp, self.indices, self.angle_deltas)
+
+        self.my_state.angle_deltas = np.roll(interp_angle_deltas, -self.next_road_tile)
+        self.my_state.reward = self.reward
+        self.my_state.on_road = self.on_road
+        self.my_state.laps = self.laps
+        self.my_state.true_speed = true_speed
+        self.my_state.off_center = off_center
+        self.my_state.wheel_angle = wheel_angle
+        self.my_state.car_angle = car_angle
+        self.my_state.angular_vel = angular_vel
+        self.my_state.vel_angle = vel_angle
+
+        # Normalization
+        self.my_state.angle_deltas *= 2.3
+        self.my_state.true_speed /= 100.0
+        self.my_state.off_center /= TRACK_WIDTH
+        self.my_state.wheel_angle *= 2.1
+        self.my_state.car_angle *= 1.5
+        self.my_state.vel_angle *= 1.5
+        self.my_state.angular_vel /= 3.74
+
+    def reset(self, *, seed: int | None = None, options: dict | None = None):
+        if seed is not None:
+            if seeding is not None:
+                self.np_random, _ = seeding.np_random(seed)
+            else:
+                self.np_random = np.random.RandomState(seed)
+        self._build_new_episode()
+        # Wichtig: initiale Features befüllen
+        self._update_features()
+        obs = self.my_state.as_feature_vector(LOOK_AHEAD).astype(np.float32)
+        info = {}
+        return obs, info
+
+    def fast_reset(self):
+        # keep the same track, respawn car
+        self.car = None
+        self.laps = 0
+        self.on_road = True
+        self.next_road_tile = 0
+
+        self.reward = 0.0
+        self.prev_reward = 0.0
+        self.tile_visited_count = 0
+        self.t = 0.0
+        self.human_render = False
+        for tile in self.road:
+            tile.road_visited = False
+        self.road_poly = [((list(poly)), list(color)) for (poly, color) in self.original_road_poly]
+        try:
+            self.car.destroy()
+        except Exception:
+            pass
+        self.car = Car(self.world, *self.track[0][1:4])
+        self.car.tiles = set()
+
+        self.steps = 0
+        self._last_progress_count = 0
+        self._no_progress_steps = 0
+        self._stall_steps = 0
+
+        return self.step(np.array([0.0, 0.0, 0.0], dtype=np.float32))
+
+    def step(self, action):
+        # log.info("got action: {}".format(action))
+        # Expect action: [steer (-1..1), gas (0..1), brake (0..1)]
+        if action is not None:
+            # TODO: this was changed! minus in steer was removed
+            self.car.steer(float(action[0]))
+            self.car.gas(float(action[1]))
+            self.car.brake(float(action[2]))
+
+        self.car.step(1.0 / FPS)
+        self.world.Step(1.0 / FPS, 6 * 30, 2 * 30)
+        self.t += 1.0 / FPS
+
+        self.steps += 1
+
+        terminated = False
+        truncated = False
+
+        if action is not None:
+            # (1) ALLE Reward-Änderungen zuerst einarbeiten
+            self.reward -= 5.0 / FPS  # Zeitstrafe
+
+            # Ziel erreicht?
+            if self.tile_visited_count == len(self.track):
+                terminated = True
+
+            # Out-of-bounds: Strafe IN reward addieren (nicht step_reward überschreiben)
+            x, y = self.car.hull.position
+            if abs(x) > PLAYFIELD or abs(y) > PLAYFIELD:
+                self.reward -= 100.0
+                terminated = True
+
+            # Offroad: kontinuierliche Strafe + Grace-Fenster; bei Timeout Zusatzstrafe
+            if not self.on_road:
+                self.offroad_frames += 1
+                self.reward -= self.offroad_penalty_per_frame / FPS
+                if self.offroad_frames > self.offroad_grace_frames:
+                    self.reward -= 20.0
+                    terminated = True
+            else:
+                self.offroad_frames = 0
+
+            if self.tile_visited_count > self._last_progress_count:
+                self._last_progress_count = self.tile_visited_count
+                self._no_progress_steps = 0
+            else:
+                self._no_progress_steps += 1
+            if self._no_progress_steps >= NO_PROGRESS_STEPS:
+                truncated = True
+
+            # (2) JETZT genau einmal das Delta bilden
+            step_reward = self.reward - self.prev_reward
+            self.prev_reward = self.reward
+        else:
+            step_reward = 0.0
+
+        # --- Feature computation (unverändert) ---
+        self._update_features()
+
+        obs = self.my_state.as_feature_vector(LOOK_AHEAD).astype(np.float32)
+        info = {}  # features nicht mehr nötig
+        return obs, step_reward, terminated, truncated, info
+
+    def _get_observation(self):
+        # This env is feature-first; return None unless user asks for rgb_array via render()
+        return None
+
+    # ------------------------
+    # Rendering (pygame)
+    # ------------------------
+    def render(self):
+        self._ensure_pygame()
+
+        # Handle window events only in human mode
+        if self.render_mode == "human":
+            for event in pygame.event.get():
+                if event.type == pygame.QUIT:
+                    self.close()
+                    return None
+
+        # Camera math (match original)
+        zoom = 0.1 * SCALE * max(1 - self.t, 0) + ZOOM * SCALE * min(self.t, 1)
+        scroll_x = self.car.hull.position[0]
+        scroll_y = self.car.hull.position[1]
+        angle = -self.car.hull.angle
+        vel = self.car.hull.linearVelocity
+        if np.linalg.norm(vel) > 0.5:
+            angle = math.atan2(vel[0], vel[1])
+
+        # Draw grass background
+        self._pg.screen.fill((102, 230, 102))
+        # simple grid for texture
+        k = PLAYFIELD / 20.0
+        grid_color = (110, 240, 110)
+        for x in range(-20, 20, 2):
+            for y in range(-20, 20, 2):
+                x0, y0 = k * x + 0, k * y + 0
+                x1, y1 = k * x + k, k * y + k
+                p0 = self._world_to_screen(x0, y0, zoom, angle, scroll_x, scroll_y)
+                p1 = self._world_to_screen(x1, y0, zoom, angle, scroll_x, scroll_y)
+                p2 = self._world_to_screen(x1, y1, zoom, angle, scroll_x, scroll_y)
+                p3 = self._world_to_screen(x0, y1, zoom, angle, scroll_x, scroll_y)
+                pygame.draw.polygon(self._pg.screen, grid_color, [p0, p1, p2, p3])
+
+        # Road polygons
+        for poly, color in self.road_poly:
+            self._draw_polygon_world(poly, color, zoom, angle, scroll_x, scroll_y)
+
+        # Draw car hull + wheels (approx)
+        car_col = (0.25, 0.25, 0.25)
+        self._draw_body(self.car.hull, car_col, zoom, angle, scroll_x, scroll_y)
+        for w in self.car.wheels:
+            self._draw_body(w, (0.15, 0.15, 0.15), zoom, angle, scroll_x, scroll_y)
+
+        # Indicators (speed, wheel, gyro)
+        if self._pg.font is not None:
+            # simple HUD text
+            txt = f"reward={self.reward:0.1f}  laps={self.laps}"
+            surf = self._pg.font.render(txt, True, (255, 255, 255))
+            self._pg.screen.blit(surf, (10, 10))
+
+        # Output
+        if self.render_mode == "human":
+            pygame.display.flip()
+            self._pg.clock.tick(FPS)
+            return None
+        else:
+            # Offscreen: return RGB array like gym does
+            arr = pygame.surfarray.array3d(self._pg.screen)  # (W,H,3)
+            arr = np.transpose(arr, (1, 0, 2))  # -> (H,W,3)
+            return arr
+
+    def close(self):
+        try:
+            if self._pg and self._pg.initialized:
+                pygame.display.quit()
+                pygame.quit()
+        except Exception:
+            pass
+        self._pg = None
+
+
+# ----------------------------
+# Keyboard demo (pygame)
+# ----------------------------
+if __name__ == "__main__":
+    import pygame
+
+    pygame.init()
+    env = CarRacing(render_mode="human")
+
+    action = np.array([0.0, 0.0, 0.0], dtype=np.float32)
+    running = True
+
+
+    def handle_keys(a):
+        keys = pygame.key.get_pressed()
+        steer = 0.0
+        if keys[pygame.K_LEFT]:
+            steer -= 1.0
+        if keys[pygame.K_RIGHT]:
+            steer += 1.0
+        gas = 1.0 if keys[pygame.K_UP] else 0.0
+        brake = 0.5 if keys[pygame.K_DOWN] else 0.0
+        a[0], a[1], a[2] = steer, gas, brake
+
+
+    # initial reset
+    env.reset()
+    try:
+        while running:
+            for event in pygame.event.get():
+                if event.type == pygame.QUIT:
+                    running = False
+                if event.type == pygame.KEYDOWN and event.key == pygame.K_RETURN:
+                    env.fast_reset()
+
+            handle_keys(action)
+            obs, r, terminated, truncated, info = env.step(action)
+
+            # print every ~200 frames
+            if int(env.t * FPS) % 200 == 0:
+                ms: MyState = info.get("features")
+                if ms is not None:
+                    print(
+                        f"speed={ms.true_speed:5.2f} off_center={ms.off_center:+.2f} car_ang={ms.car_angle:+.2f} "
+                        f"reward={r:+.2f}"
+                        f"reward={ms.reward:+.2f}"
+                    )
+
+            env.render()
+            if terminated or truncated:
+                env.fast_reset()
+
+    finally:
+        env.close()

mathema/eval_main.py

@@ -0,0 +1,48 @@
+import asyncio
+import logging
+from dotenv import load_dotenv
+
+from mathema.core.population_monitor import init_population
+from mathema.utils.logging_config import setup_logging
+
+setup_logging()
+log = logging.getLogger(__name__)
+
+N_RUNS = 10
+
+
+async def run_single_car_experiment(run_idx: int):
+    pop_id = f"car_pop_run{run_idx:02d}"
+
+    log.info(f"=== START RUN {run_idx + 1}/{N_RUNS} ({pop_id}) ===")
+
+    monitor = await init_population((
+        pop_id,
+        [{"morphology": "car_racing_features", "neural_afs": ["tanh"]}],
+        "gt",
+        "competition",
+    ))
+
+    # 👉 warten, bis der Monitor sich selbst beendet
+    await monitor._stopped_evt.wait()
+
+    # optional: letzte Stats loggen
+    s = monitor.state
+    best = await monitor._best_fitness_in_population(s.population_id)
+    log.info(
+        f"=== END RUN {run_idx + 1}/{N_RUNS} "
+        f"gens={s.pop_gen} best_fitness={best:.6f} evals={s.eval_acc} ==="
+    )
+
+
+async def main():
+    load_dotenv()
+
+    for i in range(N_RUNS):
+        await run_single_car_experiment(i)
+
+    log.info("=== ALL RUNS FINISHED ===")
+
+
+if __name__ == "__main__":
+    asyncio.run(main())

mathema/exoself_test.py

@@ -0,0 +1,21 @@
+import asyncio
+
+from mathema.core.exoself import Exoself
+from mathema.genotype.neo4j.genotype import load_genotype_snapshot
+
+
+async def main():
+    print("i am here!")
+    snapshot = await load_genotype_snapshot("08bf4d92d8c0438295399f8f2a8fef1a")
+
+    print("gathered snapshot")
+    print(snapshot)
+
+    print("------- build exoself ---------")
+    exo = Exoself(snapshot)
+
+    print("-------- building processes ---------")
+    exo.build_pid_map_and_spawn()
+
+if __name__ == "__main__":
+    asyncio.run(main())

mathema/genotype/neo4j/genotype_connector.py

@@ -0,0 +1,3 @@
+"""
+read and write api for exoself
+"""

mathema/main.py

@@ -0,0 +1,17 @@
+import asyncio
+
+from mathema.genotype.neo4j.genotype import test_add_neuron
+
+
+async def main():
+    # polis = Polis()
+    # await polis.create()
+    # await polis.start()
+
+    # await polis.stop()
+    # await test_mut_operators()
+    await test_add_neuron()
+
+
+if __name__ == '__main__':
+    asyncio.run(main())

mathema/neuron.py

@@ -1,108 +0,0 @@
-# actors/neuron.py
-import math
-import random
-
-from actor import Actor
-
-
-def tanh(x): return math.tanh(x)
-
-
-class Neuron(Actor):
-    def __init__(self, nid, cx_pid, af_name, input_idps, output_pids):
-        super().__init__(f"Neuron-{nid}")
-        self.nid = nid
-        self.cx_pid = cx_pid
-        self.af = tanh if af_name == "tanh" else tanh
-        # input_idps: [(input_id, [w1, w2, ...])]
-        self.inputs = {}
-        self.order = []
-
-        """
-        for (inp_id, weights) in input_idps:
-            self.order.append(inp_id)
-            self.inputs[inp_id] = {"weights": list(weights), "got": False, "val": None}
-        """
-
-        self.bias = 0.0
-        for (inp_id, weights) in input_idps:
-            if inp_id == "bias":
-                self.bias = float(weights[0])
-            else:
-                self.order.append(inp_id)
-                self.inputs[inp_id] = {"weights": list(weights), "got": False, "val": None}
-
-        self._backup_inputs = None
-        self._backup_bias = None
-
-        self.outputs = output_pids
-
-        print(f"Neuron {nid}: inputs={list(self.inputs.keys())}, bias={self.bias}")
-
-    async def run(self):
-        while True:
-            msg = await self.inbox.get()
-            tag = msg[0]
-
-            if tag == "forward":
-                _, from_id, data = msg
-                if from_id not in self.inputs:
-                    continue
-                slot = self.inputs[from_id]
-                slot["got"] = True
-                slot["val"] = data
-
-                if all(self.inputs[i]["got"] for i in self.order):
-                    acc = 0.0
-                    for i in self.order:
-                        w = self.inputs[i]["weights"]
-                        v = self.inputs[i]["val"]
-                        if isinstance(v, list):
-                            acc += sum(wj * vj for wj, vj in zip(w, v))
-                        else:
-                            acc += w[0] * float(v)
-                    out = self.af(acc + self.bias)
-                    for pid in self.outputs:
-                        await pid.send(("forward", self.nid, [out]))
-                    for i in self.order:
-                        self.inputs[i]["got"] = False
-                        self.inputs[i]["val"] = None
-                    print(f"Neuron {self.nid}: input_sum={acc + self.bias:.3f}, output={out:.3f}")
-
-            elif tag == "get_backup":
-                idps = [(i, self.inputs[i]["weights"]) for i in self.order]
-                idps.append(("bias", self.bias))
-                await self.cx_pid.send(("backup_from_neuron", self.nid, idps))
-
-            elif tag == "weight_backup":
-                print(f"Neuron {self.nid}: backing up weights")
-                self._backup_inputs = {k: {"weights": v["weights"][:]} for k, v in self.inputs.items()}
-                self._backup_bias = self.bias
-
-            elif tag == "weight_restore":
-                if self._backup_inputs is not None:
-                    for k in self.inputs:
-                        self.inputs[k]["weights"] = self._backup_inputs[k]["weights"][:]
-                    self.bias = self._backup_bias
-
-            elif tag == "weight_perturb":
-                print(f"Neuron {self.nid}: perturbing {len([w for i in self.order for w in self.inputs[i]['weights']])} weights")
-                tot_w = sum(len(self.inputs[i]["weights"]) for i in self.order) + 1
-                mp = 1 / math.sqrt(tot_w)
-                delta_mag = 2.0 * math.pi
-                sat_lim = 2.0 * math.pi
-
-                for i in self.order:
-                    ws = self.inputs[i]["weights"]
-                    for j in range(len(ws)):
-                        if random.random() < mp:
-                            ws[j] = _sat(ws[j] + (random.random() - 0.5) * delta_mag, -sat_lim, sat_lim)
-                if random.random() < mp:
-                    self.bias = _sat(self.bias + (random.random() - 0.5) * delta_mag, -sat_lim, sat_lim)
-
-            elif tag == "terminate":
-                return
-
-
-def _sat(val, lo, hi):
-    return lo if val < lo else (hi if val > hi else val)

mathema/population_monitor_test.py

@@ -0,0 +1,49 @@
+# tests/test_population_monitor_integration_lite.py
+import asyncio
+import random
+
+import mathema.core.population_monitor as pm
+
+
+class FakeExoself:
+    def __init__(self, agent_id, monitor):
+        self.agent_id = agent_id
+        self.monitor = monitor
+        self._task = asyncio.create_task(self._run())
+
+    async def _run(self):
+        try:
+            await asyncio.sleep(0.01)
+            seed = sum(ord(c) for c in str(self.agent_id)) % 1000
+            random.seed(seed)
+            fitness = 0.5 + random.random()
+            await self.monitor.cast(("terminated", self.agent_id, float(fitness), 4, 4, 10))
+        except asyncio.CancelledError:
+            pass
+
+    async def cast(self, msg):
+        if msg and msg[0] == "terminate":
+            if self._task:
+                self._task.cancel()
+
+
+async def fake_exoself_start(agent_id, monitor):
+    return FakeExoself(agent_id, monitor)
+
+
+pm.EXOSELF_START = fake_exoself_start  # 👉 sauberer DI-Hook
+
+
+async def main():
+    monitor = await pm.init_population(("pop_iso", pm.INIT_CONSTRAINTS, "gt", "competition"))
+
+    G = 3
+    for _ in range(G):
+        await monitor.gen_ended.wait()
+
+    # await monitor.gen_ended.wait()  # gezielt auf Generationsende warten
+    await monitor.stop("normal")
+
+
+if __name__ == "__main__":
+    asyncio.run(main())

mathema/replay.py

@@ -0,0 +1,86 @@
+import numpy as np
+import pygame
+
+from mathema.genotype.neo4j.genotype import load_genotype_snapshot, neo4j
+from viz_replay import build_policy_from_snapshot
+from mathema.envs.openai_car_racing import CarRacing
+
+
+async def _best_agent_in_population(population_id: str) -> str:
+    rows = await neo4j.read_all("""
+        MATCH (a:agent {population_id:$pid})
+        WHERE a.fitness IS NOT NULL
+        RETURN a.id AS id, toFloat(a.fitness) AS f
+        ORDER BY f DESC
+        LIMIT 1
+    """, pid=str(population_id))
+    print(rows)
+    if not rows:
+        raise RuntimeError(f"no agents found with fitness in '{population_id}'")
+    return str(rows[0]["id"])
+
+
+def _post_process_action(y: np.ndarray) -> np.ndarray:
+    y0 = float(y[0]) if y.size >= 1 else 0.0
+    y1 = float(y[1]) if y.size >= 2 else 0.0
+    y2 = float(y[2]) if y.size >= 3 else 0.0
+
+    steer = max(-1.0, min(1.0, y0))
+    gas = max(0.0, min(1.0, 0.5 * (y1 + 1.0)))
+    brake = max(0.0, min(1.0, 0.5 * (y2 + 1.0)))
+
+    return np.array([steer, gas, brake], dtype=np.float32)
+
+
+async def replay_best(population_id: str, seed: int = 5, lookahead: int = 10):
+    aid = await _best_agent_in_population(population_id)
+    snap = await load_genotype_snapshot(aid)
+    policy, I = build_policy_from_snapshot(snap)
+
+    env = CarRacing(seed_value=seed, render_mode="human")
+    _, _ = env.reset()
+    policy.reset_state()
+
+    _ = env.step(np.array([0.0, 0.0, 0.0], dtype=np.float32))
+
+    frame = 0
+
+    try:
+        while True:
+            feats = np.array(env.get_feature_vector(lookahead), dtype=np.float32)
+
+            if feats.shape[0] < I:
+                x = np.zeros((I,), dtype=np.float32)
+                x[:feats.shape[0]] = feats
+            else:
+                x = feats[:I]
+
+            y = policy.step(x)
+            act = _post_process_action(y)
+            _, r, terminated, truncated, _ = env.step(act)
+
+            if frame % 2 == 0:
+                env.render()
+
+            frame += 1
+
+            if not pygame.display.get_init() or pygame.display.get_surface() is None:
+                break
+
+            if terminated:
+                env.tile_visited_count = 0
+                env.prev_reward = env.reward
+
+                continue
+
+            if truncated:
+                env._no_progress_steps = 0
+                continue
+    finally:
+        env.close()
+
+
+if __name__ == "__main__":
+    import asyncio
+
+    asyncio.run(replay_best(population_id="car_pop", seed=1, lookahead=10))

mathema/runs/car_pop_run00__5aebb0d3/progress.csv

@@ -0,0 +1,6 @@
+t_sec,t_norm,pop_gen,best_gen,best_so_far,avg,std
+149.598299,0.083110,1,31.1378419453,31.1378419453,8.3156990881,10.4468610324
+218.266750,0.121259,2,107.0880445795,107.0880445795,13.8780192503,32.4749862952
+388.852873,0.216029,3,709.3696555218,709.3696555218,90.7517477204,210.0076645724
+474.285299,0.263492,4,708.0363221884,709.3696555218,89.0388449848,209.4723157637
+570.957412,0.317199,5,829.9438196555,829.9438196555,261.6982674772,318.9569529186

mathema/scape/car_racing.py

@@ -0,0 +1,47 @@
+import numpy as np
+import logging
+from mathema.actors.actor import Actor
+
+log = logging.getLogger(__name__)
+
+
+class CarRacingScape(Actor):
+    def __init__(self, env, name: str = "CarRacingScape"):
+        super().__init__(name)
+        self.env = env
+        self._stepped = False
+
+    def _get_features(self) -> list[float]:
+        if not self._stepped:
+            _, _, term, trunc, _ = self.env.step(np.array([0.0, 0.0, 0.0], dtype=np.float32))
+            self._stepped = True
+        return self.env.get_feature_vector()
+
+    async def run(self):
+        while True:
+            msg = await self.inbox.get()
+            tag = msg[0]
+
+            if tag == "sense":
+                _, sid, sensor_pid = msg
+                vec = self._get_features()
+                await sensor_pid.send(("percept", vec))
+
+            elif tag == "action":
+                _, action, actuator_pid = msg
+                _, step_reward, terminated, truncated, _ = self.env.step(np.asarray(action, dtype=np.float32))
+                self._stepped = True
+
+                halt_flag = 1 if (terminated or truncated) else 0
+                await actuator_pid.send(("result", float(step_reward), halt_flag))
+
+                if halt_flag == 1:
+                    self.env.fast_reset()
+                    self._stepped = False
+
+            elif tag == "terminate":
+                try:
+                    self.env.close()
+                except Exception:
+                    pass
+

mathema/scape/scape.py

@@ -1,5 +1,7 @@
-from actor import Actor
+from mathema.actors.actor import Actor
 import math
+import logging
+log = logging.getLogger(__name__)
 
 
 class XorScape(Actor):
@@ -21,11 +23,11 @@ class XorScape(Actor):
             tag = msg[0]
 
             if tag == "sense":
-                print("SCAPE: got sensed by sensor...")
+                log.debug("SCAPE: got sensed by sensor...")
                 _, sid, from_pid = msg
                 self.last_actuator = from_pid
                 inputs, correct = self.data[self.index]
-                print("SENSOR input /correct: ", inputs, correct)
+                log.debug("SENSOR input /correct: ", inputs, correct)
                 await from_pid.send(("percept", inputs))
 
             elif tag == "action":
@@ -34,7 +36,7 @@ class XorScape(Actor):
                 step_error = sum((o - c) ** 2 for o, c in zip(output, correct))
                 step_rmse = math.sqrt(step_error)
 
-                print(f"XOR PATTERN: Input={self.data[self.index][0]} → Network={output} → Expected={correct}")
+                log.debug(f"XOR PATTERN: Input={self.data[self.index][0]} → Network={output} → Expected={correct}")
 
                 self.index += 1
 

mathema/smoke_test_car.py

@@ -0,0 +1,145 @@
+import asyncio
+import logging
+import numpy as np
+
+from mathema.actors.actor import Actor
+from mathema.actors.sensor import Sensor
+from mathema.actors.actuator import Actuator
+from mathema.scape.car_racing import CarRacingScape
+from mathema.envs.openai_car_racing import CarRacing
+
+logging.basicConfig(level=logging.INFO)
+log = logging.getLogger("smoke")
+
+FEATURE_LEN = 10 + 6
+
+
+class DummyCortex(Actor):
+    """Minimaler Cortex: zählt Fitness und Episoden. Erwartet Actuator->('sync', aid, fitness, halt_flag)."""
+
+    def __init__(self, stop_after_episodes: int = 3):
+        super().__init__("DummyCortex")
+        self.total_fitness = 0.0
+        self.episodes = 0
+        self.stop_after = int(stop_after_episodes)
+        self.sensors = []
+        self.neurons = []
+        self.actuators = []
+
+    async def run(self):
+        log.info("[Cortex] started. stop_after=%d", self.stop_after)
+        try:
+            while True:
+                msg = await self.inbox.get()
+                tag = msg[0]
+                if tag == "sync":
+                    _, aid, fitness_delta, halt_flag = msg
+                    self.total_fitness += float(fitness_delta)
+                    if halt_flag == 1:
+                        self.episodes += 1
+                        log.info("[Cortex] EPISODE done: %d  cum_fitness=%.3f",
+                                 self.episodes, self.total_fitness)
+                        if self.episodes >= self.stop_after:
+                            log.info("[Cortex] stopping smoke test...")
+
+                            for a in (self.sensors + self.neurons + self.actuators):
+                                try:
+                                    await a.send(("terminate",))
+                                except Exception:
+                                    pass
+                            return
+                elif tag == "reactivate":
+
+                    pass
+                elif tag == "terminate":
+                    return
+        finally:
+            log.info("[Cortex] terminated.")
+
+
+class RelayNeuron(Actor):
+    """
+    Minimal-Neuron: nimmt Sensor-Features entgegen ("forward", sid, vec)
+    und sendet eine 3-dimensionale Aktor-Action ("forward", nid, [steer,gas,brake]) weiter.
+    """
+
+    def __init__(self, nid: str, out_actuator: Actuator):
+        super().__init__(f"RelayNeuron-{nid}")
+        self.nid = nid
+        self.out = out_actuator
+
+    async def run(self):
+        try:
+            while True:
+                msg = await self.inbox.get()
+                tag = msg[0]
+                if tag == "forward":
+                    _, _sid, features = msg
+
+                    action_vec = [0.0, 0.2, -1.0]
+                    await self.out.send(("forward", self.nid, action_vec))
+                elif tag == "terminate":
+                    return
+        finally:
+            pass
+
+
+async def main():
+    env = CarRacing(seed_value=5, render_mode=None)
+    scape = CarRacingScape(env)
+
+    cx = DummyCortex(stop_after_episodes=3)
+
+    actuator = Actuator(
+        aid="A1",
+        cx_pid=cx,
+        name="car_ApplyAction",
+        fanin_ids=["N1"],
+        expect_count=1,
+        scape=scape
+    )
+
+    neuron = RelayNeuron("N1", actuator)
+
+    sensor = Sensor(
+        sid="S1",
+        cx_pid=cx,
+        name="car_GetFeatures",
+        vector_length=FEATURE_LEN,
+        fanout_pids=[neuron],
+        scape=scape
+    )
+
+    cx.sensors = [sensor]
+    cx.neurons = [neuron]
+    cx.actuators = [actuator]
+
+    tasks = [
+        asyncio.create_task(scape.run(), name="CarScape"),
+        asyncio.create_task(cx.run(), name="Cortex"),
+        asyncio.create_task(sensor.run(), name="Sensor"),
+        asyncio.create_task(neuron.run(), name="Neuron"),
+        asyncio.create_task(actuator.run(), name="Actuator"),
+    ]
+
+    steps = 0
+    try:
+        while not tasks[1].done():
+            await sensor.send(("sync",))
+            steps += 1
+
+            await asyncio.sleep(0.0)
+        log.info("[SMOKE] finished after %d steps. ✅", steps)
+    finally:
+
+        try:
+            await scape.send(("terminate",))
+        except Exception:
+            pass
+        for t in tasks:
+            if not t.done():
+                t.cancel()
+
+
+if __name__ == "__main__":
+    asyncio.run(main())

mathema/stats/car_pop.jsonl

@@ -0,0 +1,19 @@
+{"gen":1,"ts":429466,"cum_fitness":406.1130192502508,"best":213.04574468084934,"avg":40.611301925025074,"std":81.34536458570878,"agents":10,"eval_acc":229,"cycle_acc":112776.0,"time_acc":504.0}
+{"gen":2,"ts":429516,"cum_fitness":385.63829787233476,"best":194.25435663626644,"avg":38.56382978723347,"std":77.13033019288423,"agents":10,"eval_acc":436,"cycle_acc":187892.0,"time_acc":794.0}
+{"gen":3,"ts":429604,"cum_fitness":828.4902735562338,"best":310.83920972644523,"avg":75.31729759602125,"std":124.90189226798081,"agents":11,"eval_acc":714,"cycle_acc":291121.0,"time_acc":1305.0}
+{"gen":4,"ts":429682,"cum_fitness":671.7009118541074,"best":309.0392097264455,"avg":74.63343465045638,"std":126.25620416926591,"agents":9,"eval_acc":976,"cycle_acc":410377.0,"time_acc":1694.0}
+{"gen":5,"ts":429724,"cum_fitness":893.6996453900745,"best":309.57254305977864,"avg":99.29996059889716,"std":140.72493489483733,"agents":9,"eval_acc":1157,"cycle_acc":487735.0,"time_acc":1904.0}
+{"gen":6,"ts":429770,"cum_fitness":1133.7375886524387,"best":539.5438196554759,"avg":113.37375886524387,"std":184.31148311830162,"agents":10,"eval_acc":1329,"cycle_acc":574588.0,"time_acc":2210.0}
+{"gen":7,"ts":429816,"cum_fitness":1744.1695035460898,"best":836.2938196555104,"avg":158.56086395873544,"std":243.32059311203213,"agents":11,"eval_acc":1503,"cycle_acc":670749.0,"time_acc":2500.0}
+{"gen":8,"ts":429885,"cum_fitness":2040.9778115501408,"best":875.7271529888461,"avg":226.7753123944601,"std":258.41233668126716,"agents":9,"eval_acc":1734,"cycle_acc":788501.0,"time_acc":2949.0}
+{"gen":9,"ts":429978,"cum_fitness":2287.8132725430496,"best":793.935764944267,"avg":254.20147472700552,"std":319.4685367363856,"agents":9,"eval_acc":1932,"cycle_acc":924645.0,"time_acc":3269.0}
+{"gen":10,"ts":430023,"cum_fitness":2287.209321175259,"best":868.7271529888463,"avg":254.1343690194732,"std":333.23731351153145,"agents":9,"eval_acc":2075,"cycle_acc":1016506.0,"time_acc":3434.0}
+{"gen":11,"ts":430064,"cum_fitness":2325.6576494427395,"best":868.9604863221797,"avg":232.56576494427395,"std":327.1874379309061,"agents":10,"eval_acc":2219,"cycle_acc":1111684.0,"time_acc":3659.0}
+{"gen":12,"ts":430106,"cum_fitness":2326.053140830775,"best":883.6604863221783,"avg":232.60531408307753,"std":334.03967991212033,"agents":10,"eval_acc":2414,"cycle_acc":1202188.0,"time_acc":3894.0}
+{"gen":13,"ts":430164,"cum_fitness":2811.455116514665,"best":893.8104863221778,"avg":281.1455116514665,"std":374.3884505195095,"agents":10,"eval_acc":2613,"cycle_acc":1310617.0,"time_acc":4269.0}
+{"gen":14,"ts":430214,"cum_fitness":3671.2328267476837,"best":894.4771529888444,"avg":407.9147585275204,"std":398.5607736543426,"agents":9,"eval_acc":2777,"cycle_acc":1407899.0,"time_acc":4537.0}
+{"gen":15,"ts":430283,"cum_fitness":3473.175278622045,"best":896.1271529888445,"avg":385.90836429133833,"std":431.5960755241998,"agents":9,"eval_acc":2945,"cycle_acc":1535928.0,"time_acc":4954.0}
+{"gen":16,"ts":430360,"cum_fitness":3461.7943768996524,"best":896.2271529888443,"avg":314.70857971815025,"std":416.6829267465861,"agents":11,"eval_acc":3130,"cycle_acc":1675309.0,"time_acc":5326.0}
+{"gen":17,"ts":430416,"cum_fitness":5278.096251266399,"best":898.7938196555115,"avg":439.8413542721999,"std":439.94310781767507,"agents":12,"eval_acc":3309,"cycle_acc":1797053.0,"time_acc":5697.0}
+{"gen":18,"ts":430538,"cum_fitness":5295.679584599731,"best":899.1938196555117,"avg":441.3066320499776,"std":441.3905509458581,"agents":12,"eval_acc":3611,"cycle_acc":1982624.0,"time_acc":6595.0}
+{"gen":19,"ts":430661,"cum_fitness":5304.096251266397,"best":900.5771529888449,"avg":442.0080209388664,"std":442.0866607933945,"agents":12,"eval_acc":3858,"cycle_acc":2183705.0,"time_acc":7363.0}

mathema/stats/car_pop_run00.jsonl

@@ -0,0 +1,5 @@
+{"ts":1765629719,"gen":1,"t_sec":357578,"cum_fitness":83.15699088145817,"best":31.13784194528847,"avg":8.315699088145816,"std":10.446861032430222,"agents":10,"eval_acc":273,"cycle_acc":146133.0,"time_acc":843.0}
+{"ts":1765629719,"gen":2,"t_sec":357647,"cum_fitness":138.78019250253263,"best":107.08804457953383,"avg":13.878019250253264,"std":32.4749862952049,"agents":10,"eval_acc":453,"cycle_acc":231231.0,"time_acc":1253.0}
+{"ts":1765629719,"gen":3,"t_sec":357818,"cum_fitness":907.5174772036397,"best":709.3696555217725,"avg":90.75174772036397,"std":210.0076645723655,"agents":10,"eval_acc":749,"cycle_acc":389454.0,"time_acc":2210.0}
+{"ts":1765629719,"gen":4,"t_sec":357903,"cum_fitness":890.3884498480174,"best":708.0363221884387,"avg":89.03884498480174,"std":209.47231576372903,"agents":10,"eval_acc":921,"cycle_acc":493606.0,"time_acc":2788.0}
+{"ts":1765629719,"gen":5,"t_sec":358000,"cum_fitness":2616.982674771991,"best":829.9438196555033,"avg":261.69826747719907,"std":318.9569529186433,"agents":10,"eval_acc":1135,"cycle_acc":607753.0,"time_acc":3398.0}

mathema/stats/xor_pop.jsonl

@@ -0,0 +1,45 @@
+{"gen":1,"ts":321948,"cum_fitness":3.535490412154955,"best":0.707098082430991,"avg":0.35354904121549546,"std":0.35354904121549546,"agents":10,"eval_acc":945,"cycle_acc":3780.0,"time_acc":0.0}
+{"gen":2,"ts":321949,"cum_fitness":3.5354927525295303,"best":0.7071004228055664,"avg":0.35354927525295304,"std":0.35354927525357277,"agents":10,"eval_acc":1501,"cycle_acc":6004.0,"time_acc":0.0}
+{"gen":3,"ts":321949,"cum_fitness":29.581795443265467,"best":29.581795443265467,"avg":3.6977244304081833,"std":9.783259259708018,"agents":8,"eval_acc":2169,"cycle_acc":8676.0,"time_acc":0.0}
+{"gen":4,"ts":321950,"cum_fitness":8.856365418978141,"best":6.735068112430009,"avg":0.8051241289980129,"std":1.9004936297986483,"agents":11,"eval_acc":3099,"cycle_acc":12396.0,"time_acc":0.0}
+{"gen":5,"ts":321950,"cum_fitness":13.627023608376208,"best":10.091529771445954,"avg":1.1355853006980172,"std":2.7212783264525835,"agents":12,"eval_acc":4248,"cycle_acc":16992.0,"time_acc":0.0}
+{"gen":6,"ts":321951,"cum_fitness":4.2498434127003515,"best":0.7143496834727742,"avg":0.3541536177250293,"std":0.35415877239773463,"agents":12,"eval_acc":5336,"cycle_acc":21344.0,"time_acc":0.0}
+{"gen":7,"ts":321951,"cum_fitness":4.5021926658698606,"best":0.9666988224881597,"avg":0.3751827221558217,"std":0.3813687035616592,"agents":12,"eval_acc":6296,"cycle_acc":25184.0,"time_acc":0.0}
+{"gen":8,"ts":321952,"cum_fitness":4.249843526854475,"best":0.7143496834727742,"avg":0.3541536272378729,"std":0.3541587818779628,"agents":12,"eval_acc":7060,"cycle_acc":28240.0,"time_acc":0.0}
+{"gen":9,"ts":321953,"cum_fitness":4.2563711757342135,"best":0.7143496834727742,"avg":0.3546975979778511,"std":0.3547050614807084,"agents":12,"eval_acc":7818,"cycle_acc":31272.0,"time_acc":0.0}
+{"gen":10,"ts":321954,"cum_fitness":4.256374349810292,"best":0.7143496834727742,"avg":0.354697862484191,"std":0.35470532426949464,"agents":12,"eval_acc":8693,"cycle_acc":34772.0,"time_acc":0.0}
+{"gen":11,"ts":321954,"cum_fitness":4.256374829169983,"best":0.7143496834727742,"avg":0.35469790243083194,"std":0.3547053639569165,"agents":12,"eval_acc":9591,"cycle_acc":38364.0,"time_acc":0.0}
+{"gen":12,"ts":321955,"cum_fitness":4.2563743097290745,"best":0.7143496834727742,"avg":0.3546978591440895,"std":0.3547053209510696,"agents":12,"eval_acc":10435,"cycle_acc":41740.0,"time_acc":0.0}
+{"gen":13,"ts":321955,"cum_fitness":4.2563743097290745,"best":0.7143496834727742,"avg":0.3546978591440895,"std":0.3547053209510696,"agents":12,"eval_acc":11309,"cycle_acc":45236.0,"time_acc":0.0}
+{"gen":14,"ts":321956,"cum_fitness":4.2563743097290745,"best":0.7143496834727742,"avg":0.3546978591440895,"std":0.3547053209510696,"agents":12,"eval_acc":12086,"cycle_acc":48344.0,"time_acc":0.0}
+{"gen":15,"ts":321956,"cum_fitness":9.553563169632652,"best":6.010811062944182,"avg":0.8685057426938774,"std":1.6608537151826483,"agents":11,"eval_acc":13036,"cycle_acc":52144.0,"time_acc":0.0}
+{"gen":16,"ts":321957,"cum_fitness":12.924619966862355,"best":9.381867967221527,"avg":1.0770516639051964,"std":2.5266780480098343,"agents":12,"eval_acc":13983,"cycle_acc":55932.0,"time_acc":0.0}
+{"gen":17,"ts":321958,"cum_fitness":13.727800067791875,"best":10.185044594698395,"avg":1.1439833389826564,"std":2.7468318741100606,"agents":12,"eval_acc":14954,"cycle_acc":59816.0,"time_acc":0.0}
+{"gen":18,"ts":321958,"cum_fitness":91.18998836460624,"best":65.95994946269528,"avg":9.118998836460623,"std":19.619481446836627,"agents":10,"eval_acc":16167,"cycle_acc":64668.0,"time_acc":0.0}
+{"gen":19,"ts":321959,"cum_fitness":90.0204406554163,"best":37.448365587326506,"avg":10.002271183935143,"std":13.636650609486347,"agents":9,"eval_acc":17016,"cycle_acc":68064.0,"time_acc":0.0}
+{"gen":20,"ts":321959,"cum_fitness":74.49198487354458,"best":30.40442306135378,"avg":8.27688720817162,"std":10.494831537063169,"agents":9,"eval_acc":17839,"cycle_acc":71356.0,"time_acc":0.0}
+{"gen":21,"ts":321960,"cum_fitness":80.70295993186457,"best":46.691543683144985,"avg":8.070295993186457,"std":14.146068126892585,"agents":10,"eval_acc":18627,"cycle_acc":74508.0,"time_acc":0.0}
+{"gen":22,"ts":321961,"cum_fitness":17.09561480298055,"best":10.185044594698395,"avg":1.8995127558867277,"std":3.450697192453181,"agents":9,"eval_acc":19628,"cycle_acc":78512.0,"time_acc":0.0}
+{"gen":23,"ts":321962,"cum_fitness":16.09247312003576,"best":10.185044594698395,"avg":1.7880525688928621,"std":3.493782688590394,"agents":9,"eval_acc":20543,"cycle_acc":82172.0,"time_acc":0.0}
+{"gen":24,"ts":321963,"cum_fitness":16.09247312003576,"best":10.185044594698395,"avg":1.7880525688928621,"std":3.493782688590394,"agents":9,"eval_acc":21338,"cycle_acc":85352.0,"time_acc":0.0}
+{"gen":25,"ts":321963,"cum_fitness":19.678147930304753,"best":10.185044594698395,"avg":2.1864608811449724,"std":3.471541578158789,"agents":9,"eval_acc":22280,"cycle_acc":89120.0,"time_acc":0.0}
+{"gen":26,"ts":321964,"cum_fitness":54.45134956146205,"best":38.3588764414263,"avg":6.0501499512735615,"std":11.928478608078674,"agents":9,"eval_acc":23169,"cycle_acc":92676.0,"time_acc":0.0}
+{"gen":27,"ts":321964,"cum_fitness":16.09247312003576,"best":10.185044594698395,"avg":1.7880525688928621,"std":3.493782688590394,"agents":9,"eval_acc":24074,"cycle_acc":96296.0,"time_acc":0.0}
+{"gen":28,"ts":321965,"cum_fitness":22.574146861817873,"best":10.185044594698395,"avg":2.257414686181787,"std":3.601190483978599,"agents":10,"eval_acc":25061,"cycle_acc":100244.0,"time_acc":0.0}
+{"gen":29,"ts":321965,"cum_fitness":30.36345014794506,"best":14.270977027909302,"avg":3.036345014794506,"std":5.000997100783054,"agents":10,"eval_acc":26258,"cycle_acc":105032.0,"time_acc":0.0}
+{"gen":30,"ts":321966,"cum_fitness":28.898113108697004,"best":12.805639988661246,"avg":3.2109014565218894,"std":4.828483396882332,"agents":9,"eval_acc":27117,"cycle_acc":108468.0,"time_acc":0.0}
+{"gen":31,"ts":321966,"cum_fitness":34.70951927056421,"best":12.805639988661246,"avg":3.470951927056421,"std":4.646661484076346,"agents":10,"eval_acc":28082,"cycle_acc":112328.0,"time_acc":0.0}
+{"gen":32,"ts":321967,"cum_fitness":42.1733282727597,"best":15.54737562468631,"avg":4.217332827275969,"std":5.5986512947356095,"agents":10,"eval_acc":28894,"cycle_acc":115576.0,"time_acc":0.0}
+{"gen":33,"ts":321967,"cum_fitness":40.09475832177063,"best":13.46880567369724,"avg":4.0094758321770625,"std":5.198430512503286,"agents":10,"eval_acc":29751,"cycle_acc":119004.0,"time_acc":0.0}
+{"gen":34,"ts":321968,"cum_fitness":55.10813373274723,"best":18.841892708849972,"avg":6.123125970305248,"std":7.257208686081356,"agents":9,"eval_acc":30902,"cycle_acc":123608.0,"time_acc":0.0}
+{"gen":35,"ts":321969,"cum_fitness":42.17766903843793,"best":15.547716922437663,"avg":4.217766903843793,"std":5.598841221465249,"agents":10,"eval_acc":31802,"cycle_acc":127208.0,"time_acc":0.0}
+{"gen":36,"ts":321970,"cum_fitness":47.32475411472398,"best":15.547716908753264,"avg":4.732475411472398,"std":5.421208869609718,"agents":10,"eval_acc":32595,"cycle_acc":130380.0,"time_acc":0.0}
+{"gen":37,"ts":321970,"cum_fitness":77.75220612360023,"best":45.90557988681607,"avg":7.775220612360023,"std":13.340276246021885,"agents":10,"eval_acc":33459,"cycle_acc":133836.0,"time_acc":0.0}
+{"gen":38,"ts":321971,"cum_fitness":96.42953649904732,"best":64.58291026226316,"avg":9.642953649904731,"std":18.75615829845497,"agents":10,"eval_acc":34390,"cycle_acc":137560.0,"time_acc":0.0}
+{"gen":39,"ts":321971,"cum_fitness":58.00238635155899,"best":21.858327395227583,"avg":6.444709594617666,"std":7.0718675121605665,"agents":9,"eval_acc":35288,"cycle_acc":141152.0,"time_acc":0.0}
+{"gen":40,"ts":321972,"cum_fitness":85.27003380363533,"best":55.124247641710866,"avg":8.527003380363533,"std":16.171417394577574,"agents":10,"eval_acc":36020,"cycle_acc":144080.0,"time_acc":0.0}
+{"gen":41,"ts":321972,"cum_fitness":101.3393550677986,"best":71.19356890587414,"avg":9.212668642527147,"std":20.087266316539765,"agents":11,"eval_acc":37192,"cycle_acc":148768.0,"time_acc":0.0}
+{"gen":42,"ts":321973,"cum_fitness":30.145786161924455,"best":10.533479528037628,"avg":3.0145786161924457,"std":4.611786547392002,"agents":10,"eval_acc":38259,"cycle_acc":153036.0,"time_acc":0.0}
+{"gen":43,"ts":321973,"cum_fitness":35.8846428651949,"best":10.533479528037628,"avg":3.58846428651949,"std":4.557700196628995,"agents":10,"eval_acc":39190,"cycle_acc":156760.0,"time_acc":0.0}
+{"gen":44,"ts":321974,"cum_fitness":117.94828167231363,"best":82.06363880711874,"avg":11.794828167231362,"std":23.832243221106737,"agents":10,"eval_acc":40117,"cycle_acc":160468.0,"time_acc":0.0}
+{"gen":45,"ts":321975,"cum_fitness":193.04603572221515,"best":173.43372908832833,"avg":19.304603572221517,"std":51.52202402626367,"agents":10,"eval_acc":41255,"cycle_acc":165020.0,"time_acc":0.0}

mathema/test_recurrence.py

@@ -0,0 +1,142 @@
+import asyncio
+from mathema.actors.neuron import Neuron
+from mathema.actors.actor import Actor
+
+
+class Collector(Actor):
+    def __init__(self, name="Collector"):
+        super().__init__(name)
+        self.events = []
+        self._stop = asyncio.Event()
+
+    async def run(self):
+        while True:
+            msg = await self.inbox.get()
+            tag = msg[0]
+            if tag == "forward":
+                _, from_id, vec = msg
+                self.events.append((from_id, vec))
+            elif tag == "terminate":
+                self._stop.set()
+                return
+
+
+async def start(actor):
+    return asyncio.create_task(actor.run())
+
+
+async def test_feedforward():
+    col = Collector("COL-ff")
+    N = Neuron(nid="N", cx_pid=None, af_name="tanh",
+               input_idps=[("S", [1.0], False), ("bias", [0.0], False)],
+               output_pids=[col], bias=None)
+
+    tN = await start(N)
+    tC = await start(col)
+
+    await N.send(("cycle_start",))
+
+    await N.send(("forward", "S", [1.0]))
+    await asyncio.sleep(0.01)
+
+    print("FF events:", col.events)
+
+    await col.send(("terminate",))
+    await N.send(("terminate",))
+    await asyncio.gather(tN, tC)
+
+
+async def test_lateral_nonrecurrent():
+    col = Collector("COL-lat")
+    N1 = Neuron("N1", None, "tanh",
+                [("S", [1.0], False), ("bias", [0.0], False)], [], None)
+    N2 = Neuron("N2", None, "tanh",
+                [("S", [1.0], False), ("N1", [1.0], False), ("bias", [0.0], False)], [col], None)
+
+    N1.outputs = [N2]
+
+    t1 = await start(N1)
+    t2 = await start(N2)
+    tC = await start(col)
+
+    await N1.send(("cycle_start",))
+    await N2.send(("cycle_start",))
+
+    await N1.send(("forward", "S", [1.0]))
+    await N2.send(("forward", "S", [1.0]))
+    await asyncio.sleep(0.01)
+
+    await asyncio.sleep(0.01)
+
+    print("LAT events:", col.events)
+
+    await col.send(("terminate",))
+    await N1.send(("terminate",))
+    await N2.send(("terminate",))
+    await asyncio.gather(t1, t2, tC)
+
+
+async def test_recurrent_edge():
+    col = Collector("COL-rec")
+    N1 = Neuron("N1", None, "tanh",
+                [("S", [1.0], False), ("bias", [0.0], False)], [], None)
+    N2 = Neuron("N2", None, "tanh",
+                [("S", [1.0], False), ("N1", [1.0], True), ("bias", [0.0], False)], [col], None)
+
+    N1.outputs = [N2]
+
+    t1 = await start(N1)
+    t2 = await start(N2)
+    tC = await start(col)
+
+    await N1.send(("cycle_start",))
+    await N2.send(("cycle_start",))
+
+    await N1.send(("forward", "S", [1.0]))
+    await N2.send(("forward", "S", [1.0]))
+    await asyncio.sleep(0.02)
+
+    await N1.send(("cycle_start",))
+    await N2.send(("cycle_start",))
+    await N1.send(("forward", "S", [1.0]))
+    await N2.send(("forward", "S", [1.0]))
+    await asyncio.sleep(0.02)
+
+    print("REC events:", col.events)
+
+    await col.send(("terminate",))
+    await N1.send(("terminate",))
+    await N2.send(("terminate",))
+    await asyncio.gather(t1, t2, tC)
+
+
+async def test_self_loop():
+    col = Collector("COL-self")
+    N = Neuron("N", None, "tanh",
+               [("S", [1.0], False), ("N", [1.0], True), ("bias", [0.0], False)], [], None)
+
+    N.outputs = [N, col]
+
+    tN = await start(N)
+    tC = await start(col)
+
+    await N.send(("cycle_start",))
+    await N.send(("forward", "S", [1.0]))
+    await asyncio.sleep(0.02)
+
+    await N.send(("cycle_start",))
+    await N.send(("forward", "S", [1.0]))
+    await asyncio.sleep(0.02)
+
+    print("SELF events:", col.events)
+
+    await col.send(("terminate",))
+    await N.send(("terminate",))
+    await asyncio.gather(tN, tC)
+
+
+if __name__ == "__main__":
+    asyncio.run(test_feedforward())
+    asyncio.run(test_lateral_nonrecurrent())
+    asyncio.run(test_recurrent_edge())
+    asyncio.run(test_self_loop())