mrohgalf/neuroevolution
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

initial.

Browse Source
This commit is contained in:
Morten Rohgalf
2025-09-24 09:45:17 +02:00
commit 0ad84365e4
5 changed files with 639 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

BIN
.DS_Store vendored Normal file
View File

Binary file not shown.

258
experiments/genotype.json Normal file
View File

@@ -0,0 +1,258 @@
{
"cortex": {
"id": 0.38985127736117664,
"sensor_ids": [
0.5481697495393968
],
"actuator_ids": [
0.300862433948894
],
"neuron_ids": [
0.7892266733003815,
0.08993126065168999,
0.7353679413901013,
0.7986671199471203,
0.2619274118215478,
0.1878355172703,
0.5460586524022121,
0.0325156080072021
]
},
"sensor": {
"id": 0.5481697495393968,
"name": "rng",
"vector_length": 2,
"cx_id": 0.37250552293263184,
"fanout_ids": [
0.7892266733003815,
0.08993126065168999,
0.7353679413901013,
0.7986671199471203
]
},
"actuator": {
"id": 0.300862433948894,
"name": "pts",
"vector_length": 1,
"cx_id": 0.37250552293263184,
"fanin_ids": [
0.0325156080072021
]
},
"neurons": [
{
"id": 0.7892266733003815,
"layer_index": 0,
"cx_id": 0.37250552293263184,
"activation_function": "tanh",
"input_weights": [
{
"input_id": 0.5481697495393968,
"weights": [
-0.2010596737648036,
-0.017559575650012982
]
}
],
"output_ids": [
0.11787440750922895,
0.9473188607259506,
0.9045741659912035
]
},
{
"id": 0.08993126065168999,
"layer_index": 0,
"cx_id": 0.37250552293263184,
"activation_function": "tanh",
"input_weights": [
{
"input_id": 0.5481697495393968,
"weights": [
0.49425647147649876,
0.09556856915703738
]
}
],
"output_ids": [
0.11787440750922895,
0.9473188607259506,
0.9045741659912035
]
},
{
"id": 0.7353679413901013,
"layer_index": 0,
"cx_id": 0.37250552293263184,
"activation_function": "tanh",
"input_weights": [
{
"input_id": 0.5481697495393968,
"weights": [
-0.3432015569352376,
-0.15663876804924903
]
}
],
"output_ids": [
0.11787440750922895,
0.9473188607259506,
0.9045741659912035
]
},
{
"id": 0.7986671199471203,
"layer_index": 0,
"cx_id": 0.37250552293263184,
"activation_function": "tanh",
"input_weights": [
{
"input_id": 0.5481697495393968,
"weights": [
0.44235413103542676,
-0.3014661028473905
]
}
],
"output_ids": [
0.11787440750922895,
0.9473188607259506,
0.9045741659912035
]
},
{
"id": 0.2619274118215478,
"layer_index": 1,
"cx_id": 0.37250552293263184,
"activation_function": "tanh",
"input_weights": [
{
"input_id": 0.7892266733003815,
"weights": [
0.4815260544600901
]
},
{
"input_id": 0.08993126065168999,
"weights": [
-0.34742595611872107
]
},
{
"input_id": 0.7353679413901013,
"weights": [
0.1955465593022997
]
},
{
"input_id": 0.7986671199471203,
"weights": [
0.1420046463445398
]
}
],
"output_ids": [
0.804289241732289
]
},
{
"id": 0.1878355172703,
"layer_index": 1,
"cx_id": 0.37250552293263184,
"activation_function": "tanh",
"input_weights": [
{
"input_id": 0.7892266733003815,
"weights": [
-0.26492297905129614
]
},
{
"input_id": 0.08993126065168999,
"weights": [
-0.483071194536965
]
},
{
"input_id": 0.7353679413901013,
"weights": [
0.0016581996680702371
]
},
{
"input_id": 0.7986671199471203,
"weights": [
0.47010344086613354
]
}
],
"output_ids": [
0.804289241732289
]
},
{
"id": 0.5460586524022121,
"layer_index": 1,
"cx_id": 0.37250552293263184,
"activation_function": "tanh",
"input_weights": [
{
"input_id": 0.7892266733003815,
"weights": [
0.22583100100162312
]
},
{
"input_id": 0.08993126065168999,
"weights": [
0.22079487426614341
]
},
{
"input_id": 0.7353679413901013,
"weights": [
0.3514718848950448
]
},
{
"input_id": 0.7986671199471203,
"weights": [
0.03653443587296967
]
}
],
"output_ids": [
0.804289241732289
]
},
{
"id": 0.0325156080072021,
"layer_index": 2,
"cx_id": 0.37250552293263184,
"activation_function": "tanh",
"input_weights": [
{
"input_id": 0.2619274118215478,
"weights": [
0.4397546856440344
]
},
{
"input_id": 0.1878355172703,
"weights": [
0.024524621225712195
]
},
{
"input_id": 0.5460586524022121,
"weights": [
0.19908558570374346
]
}
],
"output_ids": [
0.300862433948894
]
}
]
}

196
experiments/genotype_mapper.py Normal file
View File

@@ -0,0 +1,196 @@
import json
import random
import math
from typing import List, Dict, Tuple
# ---- Hilfsfunktionen ----
def generate_id():
"""Generiert eine eindeutige ID basierend auf Zufallszahlen."""
return random.random()
def create_neural_weights(vector_length: int):
"""Erstellt eine Liste von Gewichten für eine Verbindung."""
return [random.uniform(-0.5, 0.5) for _ in range(vector_length)]
# ---- Klassen ----
class Sensor:
def __init__(self, name: str, vector_length: int):
self.id = generate_id()
self.name = name
self.vector_length = vector_length
self.cx_id = None # Wird später hinzugefügt
self.fanout_ids = [] # Verbindungen zu Neuronen
def to_dict(self):
return {
"id": self.id,
"name": self.name,
"vector_length": self.vector_length,
"cx_id": self.cx_id,
"fanout_ids": self.fanout_ids,
}
class Actuator:
def __init__(self, name: str, vector_length: int):
self.id = generate_id()
self.name = name
self.vector_length = vector_length
self.cx_id = None # Wird später hinzugefügt
self.fanin_ids = [] # Verbindungen von Neuronen
def to_dict(self):
return {
"id": self.id,
"name": self.name,
"vector_length": self.vector_length,
"cx_id": self.cx_id,
"fanin_ids": self.fanin_ids,
}
class Neuron:
def __init__(self, layer_index: int, input_ids: List[Tuple[float, int]], output_ids: List[float], cx_id: float):
self.id = generate_id()
self.layer_index = layer_index
self.cx_id = cx_id
self.activation_function = "tanh"
self.input_weights = [
{"input_id": input_id, "weights": create_neural_weights(vector_length)}
for input_id, vector_length in input_ids
]
self.output_ids = output_ids
def to_dict(self):
return {
"id": self.id,
"layer_index": self.layer_index,
"cx_id": self.cx_id,
"activation_function": self.activation_function,
"input_weights": self.input_weights,
"output_ids": self.output_ids,
}
class Cortex:
def __init__(self, sensor_ids: List[float], actuator_ids: List[float], neuron_ids: List[float]):
self.id = generate_id()
self.sensor_ids = sensor_ids
self.actuator_ids = actuator_ids
self.neuron_ids = neuron_ids
def to_dict(self):
return {
"id": self.id,
"sensor_ids": self.sensor_ids,
"actuator_ids": self.actuator_ids,
"neuron_ids": self.neuron_ids,
}
# ---- Hauptfunktionen ----
def construct_genotype(sensor_name: str, actuator_name: str, hidden_layer_densities: List[int], file_name: str):
"""
Konstruktion eines Genotyps und Speicherung in einer JSON-Datei.
"""
# Sensor erstellen
if sensor_name == "rng":
sensor = Sensor(name="rng", vector_length=2)
else:
raise ValueError(f"System does not support a sensor by the name: {sensor_name}")
# Aktuator erstellen
if actuator_name == "pts":
actuator = Actuator(name="pts", vector_length=1)
else:
raise ValueError(f"System does not support an actuator by the name: {actuator_name}")
# Neuronenschichten erstellen
output_layer_density = actuator.vector_length
layer_densities = hidden_layer_densities + [output_layer_density]
cortex_id = generate_id()
neurons = create_neuro_layers(
cortex_id,
sensor,
actuator,
layer_densities
)
# Sensor und Aktuator mit Cortex-Informationen aktualisieren
input_layer_neurons = neurons[0]
output_layer_neurons = neurons[-1]
sensor.cx_id = cortex_id
sensor.fanout_ids = [neuron.id for neuron in input_layer_neurons]
actuator.cx_id = cortex_id
actuator.fanin_ids = [neuron.id for neuron in output_layer_neurons]
# Cortex erstellen
neuron_ids = [neuron.id for layer in neurons for neuron in layer]
cortex = Cortex(
sensor_ids=[sensor.id],
actuator_ids=[actuator.id],
neuron_ids=neuron_ids
)
# Genotyp erstellen
genotype = {
"cortex": cortex.to_dict(),
"sensor": sensor.to_dict(),
"actuator": actuator.to_dict(),
"neurons": [neuron.to_dict() for layer in neurons for neuron in layer]
}
# Genotyp in Datei speichern
with open(file_name, "w") as file:
json.dump(genotype, file, indent=4)
print(f"Genotype saved to {file_name}")
def create_neuro_layers(cortex_id: float, sensor: Sensor, actuator: Actuator, layer_densities: List[int]) -> List[
List[Neuron]]:
"""
Erstellt alle Neuronen in allen Schichten des Netzwerks.
"""
neurons = []
input_ids = [(sensor.id, sensor.vector_length)]
for layer_index, layer_density in enumerate(layer_densities):
output_ids = []
if layer_index < len(layer_densities) - 1:
# IDs der nächsten Schicht generieren
output_ids = [generate_id() for _ in range(layer_densities[layer_index + 1])]
# Neuronen für die aktuelle Schicht erstellen
layer_neurons = [
Neuron(layer_index=layer_index, input_ids=input_ids, output_ids=output_ids, cx_id=cortex_id)
for _ in range(layer_density)
]
neurons.append(layer_neurons)
# Aktuelle Schicht wird die Eingabe für die nächste Schicht
input_ids = [(neuron.id, 1) for neuron in layer_neurons]
# Letzte Schicht mit Verbindung zum Aktuator erstellen
final_layer = neurons[-1]
for neuron in final_layer:
neuron.output_ids = [actuator.id]
return neurons
# ---- Beispielverwendung ----
if __name__ == "__main__":
# Genotyp erstellen und speichern
construct_genotype(
sensor_name="rng",
actuator_name="pts",
hidden_layer_densities=[4,3], # Dichten der versteckten Schichten (z. B. 4 Neuronen in der ersten Schicht, 3 in der zweiten)
file_name="genotype.json" # Name der Datei, in der der Genotyp gespeichert wird
)

55
experiments/neuron.py Normal file
View File

@@ -0,0 +1,55 @@
import random
import math
class SimpleNeuron:
def __init__(self):
# Initialisiere Gewichte und Bias mit Zufallswerten zwischen -0.5 und 0.5
self.weights = [random.uniform(-0.5, 0.5) for _ in range(2)]
self.bias = random.uniform(-0.5, 0.5)
def dot(self, input_vector):
"""
Berechnet das Skalarprodukt zwischen dem Eingabevektor und den Gewichten,
fügt den Bias hinzu und gibt das Ergebnis zurück.
"""
acc = sum(i * w for i, w in zip(input_vector, self.weights))
return acc + self.bias
def process_input(self, input_vector):
"""
Verarbeitet den Eingabevektor, berechnet den Dot-Produkt,
wendet die Aktivierungsfunktion (tanh) an und gibt das Ergebnis zurück.
"""
print("**** Processing ****")
print(f"Input: {input_vector}")
print(f"Using Weights: {self.weights} and Bias: {self.bias}")
dot_product = self.dot(input_vector)
output = math.tanh(dot_product)
print(f"Output: {output}")
return output
def sense(in_neuron, signal):
"""
Diese Funktion überprüft, ob das Signal eine Liste der Länge 2 ist,
und sendet es an das Neuron zur Verarbeitung.
"""
if isinstance(signal, list) and len(signal) == 2:
output = in_neuron.process_input(signal)
print(f"Final Output: {output}")
else:
print("The Signal must be a list of length 2")
# Beispiel zur Verwendung:
if __name__ == "__main__":
# Erstelle ein Neuron
neuron = SimpleNeuron()
# Sende ein Signal (Eingabevektor) an das Neuron
test_signal = [0.5, -0.3] # Beispiel-Eingabe
sense(neuron, test_signal)

130
experiments/simplest_nn.py Normal file
View File

@@ -0,0 +1,130 @@
import random
import math
import threading
import queue
class Neuron(threading.Thread):
def __init__(self, weights, neuron_queue, actuator_queue):
super().__init__()
self.weights = weights
self.neuron_queue = neuron_queue
self.actuator_queue = actuator_queue
self.running = True
def dot(self, input_vector):
acc = sum(i * w for i, w in zip(input_vector, self.weights[:-1]))
return acc + self.weights[-1]
def run(self):
while self.running:
try:
message = self.neuron_queue.get(timeout=1)
if message[0] == "forward":
input_vector = message[1]
print("**** Thinking ****")
print(f"Input: {input_vector}")
print(f"With Weights: {self.weights}")
dot_product = self.dot(input_vector)
output = [math.tanh(dot_product)]
self.actuator_queue.put(("forward", output))
elif message[0] == "terminate":
self.running = False
except queue.Empty:
continue
class Sensor(threading.Thread):
def __init__(self, sensor_queue, neuron_queue):
super().__init__()
self.sensor_queue = sensor_queue
self.neuron_queue = neuron_queue
self.running = True
def run(self):
while self.running:
try:
message = self.sensor_queue.get(timeout=1)
if message == "sync":
sensory_signal = [random.uniform(0, 1), random.uniform(0, 1)]
print("**** Sensing ****")
print(f"Signal from the environment: {sensory_signal}")
self.neuron_queue.put(("forward", sensory_signal))
elif message == "terminate":
self.running = False
except queue.Empty:
continue
class Actuator(threading.Thread):
def __init__(self, actuator_queue):
super().__init__()
self.actuator_queue = actuator_queue
self.running = True
def pts(self, control_signal):
print("**** Acting ****")
print(f"Using: {control_signal} to act on the environment.")
def run(self):
while self.running:
try:
message = self.actuator_queue.get(timeout=1)
if message[0] == "forward":
control_signal = message[1]
self.pts(control_signal)
elif message[0] == "terminate":
self.running = False
except queue.Empty:
continue
class Cortex(threading.Thread):
def __init__(self, sensor_queue, neuron_queue, actuator_queue):
super().__init__()
self.sensor_queue = sensor_queue
self.neuron_queue = neuron_queue
self.actuator_queue = actuator_queue
self.running = True
def run(self):
while self.running:
command = input("amna> ").strip()
if command == "sense_think_act":
self.sensor_queue.put("sync")
elif command == "terminate":
self.sensor_queue.put("terminate")
self.neuron_queue.put(("terminate",))
self.actuator_queue.put(("terminate",))
self.running = False
else:
print("Unknown command. Please use 'sense_think_act' or 'terminate'.")
print("Cortex terminated.")
if __name__ == "__main__":
sensor_queue = queue.Queue()
neuron_queue = queue.Queue()
actuator_queue = queue.Queue()
# initialize weights (including bias)
weights = [random.uniform(-0.5, 0.5) for _ in range(2)] + [random.uniform(-0.5, 0.5)] # Zwei Gewichte + Bias
sensor = Sensor(sensor_queue, neuron_queue)
neuron = Neuron(weights, neuron_queue, actuator_queue)
actuator = Actuator(actuator_queue)
sensor.start()
neuron.start()
actuator.start()
cortex = Cortex(sensor_queue, neuron_queue, actuator_queue)
cortex.start()
cortex.join()
sensor.join()
neuron.join()
actuator.join()
print("System terminated.")