/
dqn_star.py
170 lines (149 loc) · 5.32 KB
/
dqn_star.py
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
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Created on Wed Dec 26 20:12:23 2018
@author: matan
"""
from dqn import dqn_runner, DQNetwork
import astar
import qstar
import qstar_qlearn
from q_maze import q_runner
import gym
import tensorflow as tf # Deep Learning library
import gym_maze
import pygame
from skimage import transform # Help us to preprocess the frames
from skimage.color import rgb2gray # Help us to gray our frames
import pickle
import numpy as np
def is_arr_in_list(myarr, list_arrays):
return next((True for elem in list_arrays if elem is myarr), False)
MODES = {
1: 'train',
2: 'experiment1', # test on 10 unseen mazes 10X10
3: 'experiment2', # test on 5 seen mazes 20X20
4: 'experiment3' # test on same maze with alternating goal locations
}
MODE = 4
def train(grids, env):
game = dqn_runner(grids=grids, env=env)
game.episode_render = False
game.Initialized()
game.train(name='train')
return grids
def play_1(grids, env):
model_path = './models/model_train1730.ckpt' # change this
dqn_r = dqn_runner(grids, env)
tf.reset_default_graph()
network = DQNetwork(dqn_r.state_size, dqn_r.action_size, dqn_r.learning_rate)
astar_total = []
qstar_total = []
new_grids = []
stop = 0
# for grid in grids:
for i in range(1000):
env.reset()
curr_grid = env.env.maze.grid
start = env.env.player
end = env.env.target
states_qstar, counter_qstar = qstar.qstar(env, start, end, network, model_path)
env.fixed_reset(curr_grid)
states_astar, counter_astar = astar.astar(env, start, end)
if counter_astar > 39:
print('states astar: ', counter_astar, ' states qstar: ', counter_qstar)
qstar_total.append(counter_qstar)
astar_total.append(counter_astar)
new_grids.append(env.env.maze.grid)
stop += 1
if stop == 19:
break
print('Average astar: ' + str(np.mean(astar_total)))
print('Average qstar: ' + str(np.mean(qstar_total)))
pickle.dump(grids, open("grids_20_10.p", "wb"))
return
def play_2(grids, env):
astar_total = []
qstar_total = []
dqn_total = []
i = 0
for i in range(5):
# train on grid
env.reset()
grid = env.env.maze.grid
game = dqn_runner(grid, env, fixed=True)
game.max_steps = 100000
game.total_episodes = 200
game.episode_render = True
game.learning_rate = 0.1
game.decay_rate = 0.0001
game.max_tau = 500
game.memory_size = 10000
game.Initialized()
last_model = game.train(name='exp2_grid'+str(i))
model_path = './models/model_exp2_grid'+str(i)+str(last_model)+'.ckpt'
# Use trained model on same grid for testing
tf.reset_default_graph()
network = DQNetwork(game.state_size, game.action_size, game.learning_rate)
env.fixed_reset(grid)
start = env.env.player
end = env.env.target
states_astar, counter_astar = astar.astar(env, start, end)
env.fixed_reset(grid)
states_qstar, counter_qstar = qstar.qstar(env, start, end, network, model_path)
env.fixed_reset(grid)
states_dqn, counter_dqn = game.test(env, network, model_path)
print('states astar: ', counter_astar, ' states qstar: ', counter_qstar, ' states dqn: ', counter_dqn)
astar_total.append(counter_astar)
qstar_total.append(counter_qstar)
dqn_total.append(counter_dqn)
i += 1
print('Average astar: ' + str(np.mean(astar_total)))
print('Average qstar: ' + str(np.mean(qstar_total)))
print('Average dqn: ' + str(np.mean(dqn_total)))
return
def play_3(env):
astar_total = []
qstar_total = []
q_total = []
for i in range(10):
# train on grid
env.reset()
grid = env.env.maze.grid
game = q_runner(grid, env, fixed=True)
game.train()
env.fixed_reset(grid)
start = env.env.player
end = env.env.target
print('astar')
states_astar, counter_astar = astar.astar(env, start, end)
env.fixed_reset(grid)
print('q-learn')
states_q, counter_q = game.test()
env.fixed_reset(grid)
print('qstar')
states_qstar, counter_qstar = qstar_qlearn.qstar(env, start, end, game)
print('states astar: ', counter_astar, ' states qstar: ', counter_qstar, ' states dqn: ', counter_q)
astar_total.append(counter_astar)
qstar_total.append(counter_qstar)
q_total.append(counter_q)
i += 1
print('Average astar: ' + str(np.mean(astar_total)))
print('Average qstar: ' + str(np.mean(qstar_total)))
print('Average dqn: ' + str(np.mean(q_total)))
return
if __name__ == "__main__":
grids_10 = pickle.load(open("grids_10.p", "rb"))
grids_20 = pickle.load(open("grids_10.p", "rb"))
if MODES[MODE] == "train":
env = gym.make("Maze-Arr-10x10-NormalMaze-v0")
train(grids_10, env)
elif MODES[MODE] == "experiment1":
env = gym.make("Maze-Arr-10x10-NormalMaze-v0")
play_1(grids_10, env)
elif MODES[MODE] == "experiment2":
env = gym.make("Maze-Arr-8x8-NormalMaze-v0")
play_2(grids_20, env)
elif MODES[MODE] == "experiment3":
env = gym.make("Maze-Arr-12x12-NormalMaze-v0")
play_3(env)