Snake_ML/main.py at master · samueljz/Snake_ML · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
import keras
import snakeML
import matplotlib.pyplot as plt
import numpy as np
import tensorflow as tf
import pandas as pd
import geneticAlgo
import math
import tkinter

def updateCanvas(txtbox_id, generation, n_snake, fitness, length):
    scoreScreenCanvas.itemconfig(txtbox_id, text=f"Generation: {generation}, Snake: {n_snake}, Fitness: {fitness}, Length: {length}")
    scoreScreenCanvas.update()


def getPopulationFromCsv():
    df = pd.read_csv("model/generation4.csv", sep=',', header=None)
    return np.asarray(df.values)

#Manually coded. Change it when have time
def reshapeWeights(weights, input_layer_units, hidden_layers_units, output_layer_units):

    firstIndex = 0
    lastIndex = (input_layer_units * hidden_layers_units[0])
    weight_1 = np.reshape(
        weights[firstIndex:lastIndex], (input_layer_units, hidden_layers_units[0]))
    activation_1 = np.zeros(16)  # np.asarray([stackedWeights, np.zeros(9)])
    firstIndex = lastIndex
    lastIndex = firstIndex + (hidden_layers_units[0] * output_layer_units)
    weight_3 = np.reshape(
        weights[firstIndex:lastIndex], (hidden_layers_units[0], output_layer_units))
    activation_3 = np.zeros(4)  # np.asarray([stackedWeights, np.zeros(3)])
    return np.array([weight_1, activation_1, weight_3, activation_3])

def showSnake(modelName):
    model = tf.keras.models.load_model(modelName)
    snakeML.newGameML()
    snakeML.startGameML()
    game_ended = False
    num_steps = 0
    while not game_ended:
        input_layer = snakeML.getInputLayer()
        output = model.predict(np.matrix(input_layer))
        # predict output
        output = np.argmax(output)
        len, game_ended = snakeML.nextStep(output)
        num_steps += 1
    snakeML.exit()


def calcFitness(len, num_steps):
    if len < 10:
        return math.floor(num_steps * num_steps * math.pow(2, math.floor(len)))
    else:
        fitness = num_steps * num_steps
        fitness *= math.pow(2, 10)
        fitness *= (len-9)
        return fitness

#Uncomment following line to see the best performance of a particular generation
#showSnake("model/model_generation_6_best_fitness.h5")
scoreScreenMaster = tkinter.Tk()
scoreScreenCanvas = tkinter.Canvas(
    scoreScreenMaster, bg="white", height=200, width=500)
scoreScreenCanvas.pack()
txtbox_trained_id = scoreScreenCanvas.create_text(250, 20, text=f"Generation: 0, Snake: 0, Fitness: 0, Length: 0")
txtbox_training_id = scoreScreenCanvas.create_text(250, 120, text=f"Generation: 0, Snake: 0, Fitness: training")

GeneticAlgo = geneticAlgo.GeneticAlgo()
num_of_gens = 20
n_population = 2000
max_steps = 2500

# Build model (Building fixed NN first)
input_layer_units = 24
hidden_layers_units = [16]
output_layer_units = 4
model = tf.keras.models.Sequential()
model.add(tf.keras.layers.Dense(
    hidden_layers_units[0], input_dim=input_layer_units, activation=tf.nn.sigmoid))
model.add(tf.keras.layers.Dense(output_layer_units, activation=tf.nn.softmax))
total_weights = input_layer_units * hidden_layers_units[0]
total_weights += hidden_layers_units[0] * output_layer_units

# Genetic loop
population = GeneticAlgo.generatePopulation(
   n_population, total_weights)  # Change to total number of weights
#population = getPopulationFromCsv()
for i in range(1, num_of_gens):
    print(f"Generation {i}:")
    fitness = np.empty(n_population)
    len = np.empty(n_population)
    for j in range(0, n_population):
        weights = reshapeWeights(
            population[j], input_layer_units, hidden_layers_units, output_layer_units)
        model.set_weights(weights)
        snakeML.newGameML()
        snakeML.startGameML(showGui=False)
        game_ended = False
        num_steps = 0
        while not game_ended and (num_steps < max_steps):
            input_layer = snakeML.getInputLayer()
            output = model.predict(np.matrix(input_layer))
            # predict output
            output = np.argmax(output)
            len[j], game_ended = snakeML.nextStep(output)
            num_steps += 1
        fitness[j] = calcFitness(len[j], num_steps)
        snakeML.exit()
        updateCanvas(txtbox_training_id, i, j, fitness[j], len[j])
    best_person = np.argmax(fitness)
    print(f"Generation {i}, Top fitness = {fitness[best_person]}, Length = {len[j]}")
    updateCanvas(txtbox_trained_id, i, "Null", fitness[best_person], len[j])

    weights = reshapeWeights(
        population[best_person], input_layer_units, hidden_layers_units, output_layer_units)
    model.set_weights(weights)
    model.save(f"model/model_generation_{i}_best_fitness.h5")
    print(f"Generation {i} best model saved.")
    print("Creating next generation. This might take a while.")
    population = GeneticAlgo.getNextGeneration(population, fitness)
    df = pd.DataFrame(population)
    df.to_csv(f'model/generation{i+1}.csv', header=None, index=None)