forked from yao62995/ALE_dqn
/
ale_learning.py
258 lines (234 loc) · 10.3 KB
/
ale_learning.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
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
#!/usr/bin/python
# -*- coding: utf-8 -*-
# author: <yao62995@gmail.com>
import os
import random
import argparse
import time
import json
import numpy as np
import cv2
from collections import deque
import pygame
from ale_util import logger
from ale_net import DLNetwork
from ale_interface import AleInterface
pygame.init()
class DQNLearning(object):
def __init__(self, game_name, args):
self.game_name = game_name
self.logger = logger
self.game = AleInterface(game_name, args)
self.actions = self.game.get_actions_num()
# DQN parameters
self.observe = args.observe
self.explore = args.explore
self.replay_memory = args.replay_memory
self.batch_size = args.batch_size
self.gamma = args.gamma
self.init_epsilon = args.init_epsilon
self.final_epsilon = args.final_epsilon
self.save_model_freq = args.save_model_freq
self.update_frequency = args.update_frequency
self.action_repeat = args.action_repeat
self.frame_seq_num = args.frame_seq_num
if args.saved_model_dir == "":
self.param_file = "./saved_networks/%s.json" % game_name
else:
self.param_file = "%s/%s.json" % (args.saved_model_dir, game_name)
self.net = DLNetwork(game_name, self.actions, args)
# screen parameters
# self.screen = (args.screen_width, args.screen_height)
# pygame.display.set_caption(game_name)
# self.display = pygame.display.set_mode(self.screen)
self.deque = deque()
def param_serierlize(self, epsilon, step):
json.dump({"epsilon": epsilon, "step": step}, open(self.param_file, "w"))
def param_unserierlize(self):
if os.path.exists(self.param_file):
jd = json.load(open(self.param_file, 'r'))
return jd['epsilon'], jd["step"]
else:
return self.init_epsilon, 0
def process_snapshot(self, snap_shot):
# rgb to gray, and resize
snap_shot = cv2.cvtColor(cv2.resize(snap_shot, (80, 80)), cv2.COLOR_BGR2GRAY)
# image binary
# _, snap_shot = cv2.threshold(snap_shot, 1, 255, cv2.THRESH_BINARY)
return snap_shot
def show_screen(self, np_array):
return
# np_array = cv2.resize(np_array, self.screen)
# surface = pygame.surfarray.make_surface(np_array)
# surface = pygame.transform.rotate(surface, 270)
# rect = pygame.draw.rect(self.display, (255, 255, 255), (0, 0, self.screen[0], self.screen[1]))
# self.display.blit(surface, rect)
# pygame.display.update()
def train_net(self):
# training
max_reward = 0
epsilon, global_step = self.param_unserierlize()
step = 0
epoch = 0
while True: # loop epochs
epoch += 1
# initial state
self.game.reset_game()
# initial state sequences
state_seq = np.empty((80, 80, self.frame_seq_num))
for i in range(self.frame_seq_num):
state = self.game.get_screen_rgb()
self.show_screen(state)
state = self.process_snapshot(state)
state_seq[:, :, i] = state
stage_reward = 0
while True: # loop game frames
# select action
best_act = self.net.predict([state_seq])[0]
if random.random() <= epsilon or len(np.unique(best_act)) == 1: # random select
action = random.randint(0, self.actions - 1)
else:
action = np.argmax(best_act)
# carry out selected action
reward_n = self.game.act(action)
state_n = self.game.get_screen_rgb()
self.show_screen(state)
state_n = self.process_snapshot(state_n)
state_n = np.reshape(state_n, (80, 80, 1))
state_seq_n = np.append(state_n, state_seq[:, :, : (self.frame_seq_num - 1)], axis=2)
terminal_n = self.game.game_over()
# scale down epsilon
if step > self.observe and epsilon > self.final_epsilon:
epsilon -= (self.init_epsilon - self.final_epsilon) / self.explore
# store experience
act_onehot = np.zeros(self.actions)
act_onehot[action] = 1
self.deque.append((state_seq, act_onehot, reward_n, state_seq_n, terminal_n))
if len(self.deque) > self.replay_memory:
self.deque.popleft()
# minibatch train
if step > self.observe and step % self.update_frequency == 0:
for _ in xrange(self.action_repeat):
mini_batch = random.sample(self.deque, self.batch_size)
batch_state_seq = [item[0] for item in mini_batch]
batch_action = [item[1] for item in mini_batch]
batch_reward = [item[2] for item in mini_batch]
batch_state_seq_n = [item[3] for item in mini_batch]
batch_terminal = [item[4] for item in mini_batch]
# predict
target_rewards = []
batch_pred_act_n = self.net.predict(batch_state_seq_n)
for i in xrange(len(mini_batch)):
if batch_terminal[i]:
t_r = batch_reward[i]
else:
t_r = batch_reward[i] + self.gamma * np.max(batch_pred_act_n[i])
target_rewards.append(t_r)
# train Q network
self.net.fit(batch_state_seq, batch_action, target_rewards)
# update state
state_seq = state_seq_n
step += 1
# serierlize param
# save network model
if step % self.save_model_freq == 0:
global_step += step
self.param_serierlize(epsilon, global_step)
self.net.save_model("%s-dqn" % self.game_name, global_step=global_step)
self.logger.info("save network model, global_step=%d, cur_step=%d" % (global_step, step))
# state description
if step < self.observe:
state_desc = "observe"
elif epsilon > self.final_epsilon:
state_desc = "explore"
else:
state_desc = "train"
# record reward
print "game running, step=%d, action=%s, reward=%d, max_Q=%.6f, min_Q=%.6f" % \
(step, action, reward_n, np.max(best_act), np.min(best_act))
if reward_n > stage_reward:
stage_reward = reward_n
if terminal_n:
break
# record reward
if stage_reward > max_reward:
max_reward = stage_reward
self.logger.info(
"epoch=%d, state=%s, step=%d(%d), max_reward=%d, epsilon=%.5f, reward=%d, max_Q=%.6f" %
(epoch, state_desc, step, global_step, max_reward, epsilon, stage_reward, np.max(best_act)))
def play_game(self, epsilon):
# play games
max_reward = 0
epoch = 0
if epsilon == 0.0:
epsilon, _ = self.param_unserierlize()
while True: # epoch
epoch += 1
self.logger.info("game start...")
# init state
self.game.reset_game()
state_seq = np.empty((80, 80, self.frame_seq_num))
for i in range(self.frame_seq_num):
state = self.game.get_screen_rgb()
self.show_screen(state)
state = self.process_snapshot(state)
state_seq[:, :, i] = state
stage_step = 0
stage_reward = 0
while True:
# select action
best_act = self.net.predict([state_seq])[0]
if random.random() < epsilon or len(np.unique(best_act)) == 1:
action = random.randint(0, self.actions - 1)
else:
action = np.argmax(best_act)
# carry out selected action
reward_n = self.game.act(action)
state_n = self.game.get_screen_rgb()
self.show_screen(state_n)
state_n = self.process_snapshot(state_n)
state_n = np.reshape(state_n, (80, 80, 1))
state_seq_n = np.append(state_n, state_seq[:, :, : (self.frame_seq_num - 1)], axis=2)
terminal_n = self.game.game_over()
state_seq = state_seq_n
# record
if reward_n > stage_reward:
stage_reward = reward_n
if terminal_n:
time.sleep(2)
break
else:
stage_step += 1
stage_reward = reward_n
print "game running, step=%d, action=%d, reward=%d" % \
(stage_step, action, reward_n)
# record reward
if stage_reward > max_reward:
max_reward = stage_reward
self.logger.info("game over, epoch=%d, step=%d, reward=%d, max_reward=%d" %
(epoch, stage_step, stage_reward, max_reward))
def parser_argument():
parse = argparse.ArgumentParser()
parse.add_argument("--play", action="store", help="play games, you can specify model file in model directory")
parse.add_argument("--train", action="store", help="train DQNetwork, game names is needed")
parse.add_argument("--gpu", action="store", default=0, help="specify gpu number")
args = parse.parse_args()
gpu = int(args.gpu)
if args.play is not None:
if not args.play.isdigit():
game_name = args.play
else:
game_name = "breakout"
dqn = DQNLearning(game_name, gpu=None)
dqn.play_game()
elif args.train is not None:
if not args.train.isdigit():
game_name = args.train
else:
game_name = "breakout"
dqn = DQNLearning(game_name, gpu=gpu)
dqn.train_net()
else:
parse.print_help()
if __name__ == "__main__":
parser_argument()