update: horses now probably learn

genetic_alg.py

@@ -5,9 +5,9 @@ import numpy as np
 class NeuralNetwork(nn.Module):
     def __init__(self, inputSize):
         super().__init__()
-        self.hidden1 = nn.Linear(inputSize, 32)
-        self.hidden2 = nn.Linear(32, 16)
-        self.output = nn.Linear(16, 3)
+        self.hidden1 = nn.Linear(inputSize, 128)
+        self.hidden2 = nn.Linear(128, 128)
+        self.output = nn.Linear(128, 3)
 
     def forward(self, x):
         x = torch.relu(self.hidden1(x))
@@ -16,20 +16,25 @@ class NeuralNetwork(nn.Module):
         return torch.argmax(x)
 
 class GeneticAlgorithm:
-    def __init__(self, populationSize: int, mutationRate: float, percentageBest: float, inputSize: int = 5):
+    def __init__(self, populationSize: int, mutationRate: float, percentageBest: float, percentageNew: float, inputSize: int = 5):
         self.populationSize = populationSize
         self.mutationRate = mutationRate
         self.percentageBest = percentageBest
+        self.percentageNew = percentageNew
         self.inputSize = inputSize
         self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+        print(self.device)
         self.initialize_population()
 
+    def create_ind(self):
+        return NeuralNetwork(self.inputSize).to(self.device)
+
     def initialize_population(self):
-        self.population = [NeuralNetwork(self.inputSize).to(self.device) for _ in range(self.populationSize)]
+        self.population = [self.create_ind() for _ in range(self.populationSize)]
     
     def crossover(self, parent1: NeuralNetwork, parent2: NeuralNetwork):
-        child1 = NeuralNetwork(self.inputSize).to(self.device)
-        child2 = NeuralNetwork(self.inputSize).to(self.device)
+        child1 = self.create_ind()
+        child2 = self.create_ind()
         point1 = len(child1.hidden1.weight.data) // 2
         point2 = len(child1.hidden2.weight.data) // 2
         child1.hidden1.weight.data = torch.cat((parent1.hidden1.weight.data[:point1], parent2.hidden1.weight.data[point1:]), dim=0)
@@ -44,21 +49,23 @@ class GeneticAlgorithm:
     def mutate(self, model: NeuralNetwork):
         for param in model.parameters():
             if torch.rand(1).item() < self.mutationRate:
-                param.data += torch.randn_like(param.data) * 0.1
+                param.data += torch.randn_like(param.data) * 0.1 * (1 if torch.rand(1).item() >= 0.5 else -1)
         return model
 
     def learn(self, fitness: list):
-        self.population = [self.population[x] for x in np.argsort(fitness)]
+        self.population = [self.population[x] for x in np.argsort(fitness)[::-1]]
         numBest = int(self.populationSize * self.percentageBest)
         self.population = self.population[:numBest]
-        while len(self.population) < self.populationSize:
+        while len(self.population) < self.populationSize - self.populationSize * self.percentageNew:
             parent1, parent2 = np.random.choice(self.population), np.random.choice(self.population)
             child1, child2 = self.crossover(parent1, parent2)
             child1 = self.mutate(child1)
             child2 = self.mutate(child2)
             self.population.extend([child1, child2])
+        while len(self.population) < self.populationSize: self.population.append(self.create_ind())
         while len(self.population) > self.populationSize: self.population.pop()
     
     def predict(self, data: list, i):
         data = torch.tensor(data, requires_grad=False).float().to(self.device)
-        return self.population[i](data)
+        return self.population[i](data)
+