Answers.py

# These are the answer for the Flappy Bird Project. It includes some commented solutions for different possible implementations.

import random
import torch
import torch.nn as nn

MUTATION_RATE = 0.2  # Probability of each weight/bias being mutated
MUTATION_STRENGTH = 0.3
N = 10 # Number of birds to select for next round

class Model(nn.Module):
    def __init__(self, in_channels, out_channels):
        super(Model, self).__init__()
        self.fc1 = nn.Linear(in_channels, out_channels)

    def forward(self, input):
        x = self.fc1(input)
        x = torch.sigmoid(x)
        return x
  
    
# Mutates the weights and bias of a bird's model using a mutation rate and strength
# specified by the parameters
def mutate_bird(bird):
    new_bird = bird.clone()
    with torch.no_grad():
        for param in new_bird.model.parameters():
            # Sets the mutation to be between [-strength, strength]
            mutation = 2 * (torch.rand_like(param.data) - 0.5) * MUTATION_STRENGTH
            # Sets (1 - rate) of the values to be 0
            mutation = mutation * (torch.rand_like(param.data) < MUTATION_RATE).float()
            # Adds the mutation
            param.data += mutation
            
    return new_bird

"""def select_next_round(birds):
    # Sorts birds in descending order
    birds.sort(key=lambda bird: bird.deathTime, reverse=True)
    # Gets first 3 birds
    best_birds = birds[:3]
    # Prints best 3 birds' death times
    print([best_birds[i].deathTime for i in range(3)])
    return best_birds"""

# ORRRRRR
    
def select_next_round(birds):
    best_birds = []
    total_fitness = sum(bird.deathTime for bird in birds)

    if total_fitness == 0:
        return birds[:N]
    raw_weights = [bird.deathTime for bird in birds]
    
    # pick top N birds based on probabilities
    # probabilities = [fitness / total_fitness for fitness in raw_weights]
    # for _ in range(N):
    #     random_value = random.random()
    #     for i in range(len(birds)):
    #         random_value -= probabilities[i]
    #         if random_value <= 0:
    #             best_birds.append(birds[i])
    #             break
    
    # ORRRRR
    best_birds = random.choices(birds, weights=raw_weights, k=N)

    return best_birds