evolution/vis_live.py at master · nhamil/evolution · 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
124
125
126
127
128
129
130
131
132
133
import neat

import matplotlib.pyplot as plt
import numpy as np
import atari_py
import gym

import multiprocessing as mp
import sys

env = gym.make('CarRacing-v0')

# print('Input:', env.reset().shape)
# print('Output:', env.action_space)

# for x in gym.envs.registry.all():
#     print(x)

# sys.exit(0)

plt.ion()

def fitness_car_race(net: neat.Network, render: bool=False, steps=1000):
    score = 0

    for _ in range(3):
        # env._max_episode_steps = steps
        obs = env.reset()

        net.clear()

        s = 0

        while True:
            close = False

            if render:
                close = not env.render()
                # print(obs)

            obs = obs / 255.0
            obs_in = obs[::8,::8]

            # if render:
            #     plt.cla()
            #     plt.imshow(obs_in[:,:,1])
            #     plt.pause(0.00001)

            res = net.predict(obs_in.flatten())
            res = res * 2 - 1
            action = res #np.argmax(res)

            obs, reward, done, _ = env.step(action)

            s += reward

            if done or close:
                break

        score += s

        if render:
            print(s)

        env.close()

        if close:
            break

    return score / 3

if __name__ == "__main__":
    neat_args = {
        'n_pop': 100,
        'max_species': 30,
        'species_threshold': 1.0,
        'survive_threshold': 0.5,
        'clear_species': 100,
        'prob_add_node': 0.01,
        'prob_add_conn': 0.05,
        'prob_replace_weight': 0.01,
        'prob_mutate_weight': 0.5,
        'prob_toggle_conn': 0.01,
        'prob_replace_activation': 0.1,
        'std_new': 1.0,
        'std_mutate': 0.01,
        'activations': ['sigmoid'],
        'dist_weight': 0.5,
        'dist_activation': 1.0,
        'dist_disjoint': 1.0
    }

    n = neat.Neat(96//8*96//8*1, 3, neat_args)

    pool = mp.Pool()

    LENGTH = 1000
    times = 0
    best = -float('inf')

    try:
        for i in range(1000):
            scores = []
            pop = n.ask()

            for ind in pop:
                # scores.append(fitness_car_race(ind, render=True, steps=LENGTH))
                scores.append(pool.apply_async(fitness_car_race, ((ind, False, LENGTH))))

            scores = [s.get() for s in scores]

            n.tell(scores)

            max_score = np.max(scores)
            if True:
                if max_score > best:
                    best = max_score

                ind = pop[np.argmax(scores)]
                # print(ind)
                fitness_car_race(ind, render=True)

            if max_score >= LENGTH:
                times += 1
            else:
                times = 0

            if times == 5:
                print(ind)
                break

    except Exception as e:
        print("Error while training:", e)