-
Notifications
You must be signed in to change notification settings - Fork 4
/
a3c.py
326 lines (266 loc) · 11.5 KB
/
a3c.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
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
# -*- coding: utf-8 -*-
import os
import random
import threading
import numpy as np
import tensorflow as tf
import default.cartpole as cartpole
import default.mountaincar as mountaincar
from collections import deque
#Shared global hyper-parameters
T = 0 # Global shared counter
TMAX = 5000000 # Max iteration of global shared counter
THREADS = 12 # Number of running thread
N_STEP = 7 # Number of steps before update
GAMMA = 0.99 # Decay rate of past observations
ENTROPY_REGU = 0.001 # Entropy regulation term: beta, default: 0.001
INIT_LEARNING_RATE = 0.0001 # Initial learning rate
# Optimizer
OPT_DECAY = 0.99 # Discouting factor for the gradient, default: 0.99
OPT_MOMENTUM = 0.0 # A scalar tensor, default: 0.0
OPT_EPSILON = 0.005 # value to avoid zero denominator, default: 0.01
ENVIRONMENT = 1
if ENVIRONMENT == 1:
ACTIONS = 2
STATES = 4
MAX_SCORE = 3000
CHECKPOINT_DIR = 'default/save_a3c/cartpole'
elif ENVIRONMENT == 2:
ACTIONS = 3
STATES = 2
MAX_SCORE = 10
CHECKPOINT_DIR = 'default/save_a3c/mountaincar'
class netCreator(object):
def __init__(self):
with tf.device("/cpu:0"):
# Placeholder
self.s = tf.placeholder("float", [None, STATES]) # Make as input layer of network
# Network weights
self.W_fc1 = self._weight_variable([STATES, 300])
self.b_fc1 = self._bias_variable([300])
self.W_fc2 = self._weight_variable([300, 200])
self.b_fc2 = self._bias_variable([200])
self.W_fc3 = self._weight_variable([200, 100])
self.b_fc3 = self._bias_variable([100])
# weight for policy output layer
self.W_fc4 = self._weight_variable([100, ACTIONS])
self.b_fc4 = self._bias_variable([ACTIONS])
# weight for value output layer
self.W_fc5 = self._weight_variable([100, 1])
self.b_fc5 = self._bias_variable([1])
# region make fc hiddent layers
h_fc1 = tf.nn.relu(tf.matmul(self.s, self.W_fc1) + self.b_fc1)
h_fc2 = tf.nn.relu(tf.matmul(h_fc1, self.W_fc2) + self.b_fc2)
h_fc3 = tf.nn.relu(tf.matmul(h_fc2, self.W_fc3) + self.b_fc3)
# endregion make fc hiddent layers
# region output layer
# policy
self.pi = tf.nn.softmax(tf.matmul(h_fc3, self.W_fc4) + self.b_fc4)
self.log_pi = tf.log(self.pi)
self.argmax_pi = tf.argmax(self.pi, dimension=1)
# value
self.v = tf.matmul(h_fc3, self.W_fc5) + self.b_fc5
# endregion output layer
def loss_func(self):
#global ENTROPY_REGU, OPT_DECAY, OPT_MOMENTUM, OPT_EPSILON
with tf.device("/cpu:0"):
# Entropy loss
entropy = -tf.reduce_sum(self.pi * self.log_pi)
# Policy loss
self.a = tf.placeholder("float", [None, ACTIONS])
self.diff = tf.placeholder("float", [None])
self.pi_loss = -(tf.reduce_sum(tf.mul(self.log_pi, self.a), 1) * self.diff + ENTROPY_REGU * entropy)
# Value loss
self.r = tf.placeholder("float", [None])
self.v_loss = tf.reduce_mean(tf.square(self.r - self.v))
# Optimizer
self.lr = tf.placeholder("float")
self.optimizer = tf.train.RMSPropOptimizer(self.lr, OPT_DECAY, OPT_MOMENTUM, OPT_EPSILON)
self.opt_v = self.optimizer.minimize(self.v_loss)
self.opt_pi = self.optimizer.minimize(self.pi_loss) # TODO: wait.. should we maximize?
def _weight_variable(self, shape):
initial = tf.truncated_normal(shape, stddev = 0.01)
return tf.Variable(initial)
def _bias_variable(self, shape):
initial = tf.constant(0.01, shape = shape)
return tf.Variable(initial)
def forward_policy(self, sess, s_t):
pi_out = sess.run(self.pi, feed_dict = {self.s : [s_t]})
return pi_out[0]
def forward_value(self, sess, s_t):
v_out = sess.run(self.v, feed_dict = {self.s : [s_t]})
return v_out[0][0] # output is scalar
def sync_from(self, src_netowrk, name=None):
src_policy_vars = src_netowrk.get_policy_param()
src_value_vars = src_netowrk.get_value_param()
dst_policy_vars = self.get_policy_param()
dst_value_vars = self.get_value_param()
sync_ops = []
with tf.device("/cpu:0"):
with tf.op_scope([], name, "netCreator") as name:
for(src_policy_var, dst_policy_var) in zip(src_policy_vars, dst_policy_vars):
sync_op = tf.assign(dst_policy_var, src_policy_var)
sync_ops.append(sync_op)
for(src_value_var, dst_value_var) in zip(src_value_vars, dst_value_vars):
sync_op = tf.assign(dst_value_var, src_value_var)
sync_ops.append(sync_op)
return tf.group(*sync_ops, name=name)
def get_policy_param(self):
return [self.W_fc1, self.b_fc1,
self.W_fc2, self.b_fc2,
self.W_fc3, self.b_fc3,
self.W_fc4, self.b_fc4]
def get_value_param(self):
return [self.W_fc1, self.b_fc1,
self.W_fc2, self.b_fc2,
self.W_fc3, self.b_fc3,
self.W_fc5, self.b_fc5]
def train(self, sess, states, actions, R, td, learningRate):
with tf.device("/gpu:0"):
sess.run(self.opt_v, feed_dict = {self.s: states, self.r: R, self.lr: learningRate})
sess.run(self.opt_pi, feed_dict = {self.s: states, self.a: actions, self.diff: td, self.lr: learningRate})
def predict(self, sess, states):
feed_dict = {self.s: states}
a, V = sess.run([self.argmax_policy, self.V], feed_dict)
return a, V
class trainThread(object):
def __init__(self, num, global_network, initial_lr, environment):
print "THREAD ", num, "STARTING...", "LEARNING POLICY => INITIAL_LEARNING_RATE:", initial_lr
self.num = num
self.thread_network = netCreator()
self.thread_network.loss_func()
self.sync_net = self.thread_network.sync_from(global_network)
# Open communicate with environment
self.environment = environment
if self.environment == 1:
self.env_state = cartpole.CartPole()
elif self.environment == 2:
self.env_state = mountaincar.MountainCarND()
self.env_state._reset()
self.initial_lr = initial_lr
self.t = 1
self.step_score = 0
def choose_action(self, pi_values):
totals = []
running_total = 0.0
for rate in pi_values:
running_total += rate
value = running_total
totals.append(value)
rnd = random.random() * running_total
for i in range(len(totals)):
if totals[i] >= rnd:
return i
# fail safe
return np.argmax(pi_values)
def actorLearner(self, sess):
global T
# Reset gradients
states = []
actions = []
rewards = []
values = []
R = []
td = []
sess.run(self.sync_net) # Sync with global network
# Grab a state
s_t, terminal = self.env_state._state()
t_start = self.t
# Act until terminal or we did 'n_step' steps
for i in range(N_STEP):
pi_ = self.thread_network.forward_policy(sess, s_t)
v_t = self.thread_network.forward_value(sess, s_t)
# Action selection based on softmax
a_index = self.choose_action(pi_)
a_t = np.zeros([ACTIONS])
a_t[a_index] = 1
# Run one step of sim
s_t1, r_t, terminal = self.env_state._step(a_index)
# Accumulate gradients
states.append(s_t)
actions.append(a_t)
rewards.append(r_t)
values.append(v_t)
# Update counters
if self.environment == 1:
self.step_score += r_t
elif self.environment == 2:
self.step_score += 0
self.t += 1
T += 1
s_t = s_t1
if (self.num == 0) and (self.t % 100) == 0:
print "P:", pi_, "/ V", v_t, "/ ACTION", a_t
if terminal:
print "THREAD:", self.num, "/ TIME", T, "/ TIMESTEP", self.t, "/ SCORE", self.step_score
if self.environment == 1:
self.step_score = 0
elif self.environment == 2:
self.step_score += 1
# Reset state
self.env_state._reset()
break
# bootstrap if last state not terminal
R_t = 0.0 if terminal else self.thread_network.forward_value(sess, s_t)
states.reverse()
rewards.reverse()
actions.reverse()
values.reverse()
steps_done = self.t - t_start
for i in range(steps_done): # [t-1 ..., t_start] but shifted to start at 0
R.append(rewards[i] + GAMMA * R_t)
td.append(R[i] - values[i])
cur_lr = self._anneal_learning_rate(T)
return self.step_score, states, actions, R, td, cur_lr
def _anneal_learning_rate(self, global_t):
global TMAX
learning_rate = self.initial_lr * (TMAX - global_t) /TMAX
if learning_rate < 0.0:
learning_rate = 0.0
return learning_rate
def igniter(thread_index, coord, lock):
global T, TMAX, MAX_SCORE
training_thread = threads_checkin[thread_index]
score = 0
while not coord.should_stop():
if T > TMAX:
coord.request_stop()
if score > MAX_SCORE:
coord.request_stop()
score, states, actions, R, td, cur_lr = training_thread.actorLearner(sess)
lock.acquire()
global_network.train(sess, states, actions, R, td, cur_lr)
lock.release()
# Globally shared network
global_network = netCreator()
global_network.loss_func()
# Global experiences for reply
experiences = deque()
lock = threading.Lock()
threads_checkin = list()
for i in range(THREADS):
training_threads = trainThread(i, global_network, INIT_LEARNING_RATE, ENVIRONMENT)
threads_checkin.append(training_threads)
# Initialize session and variables
sess = tf.Session(config=tf.ConfigProto(log_device_placement=False))
saver = tf.train.Saver()
sess.run(tf.initialize_all_variables())
coord = tf.train.Coordinator()
checkpoint = tf.train.get_checkpoint_state(CHECKPOINT_DIR)
if checkpoint and checkpoint.model_checkpoint_path:
saver.restore(sess, checkpoint.model_checkpoint_path)
print "Successfully loaded:", checkpoint.model_checkpoint_path
if __name__ == "__main__":
# Start n concurrent actor threads
threads = list()
for i in range(THREADS):
t = threading.Thread(target=igniter, args=(i, coord, lock,))
threads.append(t)
# Start all threads
for x in threads:
x.start()
# Wait for all of them to finish
coord.join(threads)
if not os.path.exists(CHECKPOINT_DIR):
os.mkdir(CHECKPOINT_DIR)
saver.save(sess, CHECKPOINT_DIR + '/' + 'checkpoint', global_step = T)