-
Notifications
You must be signed in to change notification settings - Fork 0
/
pong_pg_fc.py
194 lines (172 loc) · 6.21 KB
/
pong_pg_fc.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
import tensorflow as tf
import numpy as np
import gym
from dl_utils import _fc
global IM_SIZE
global DIS_FACTOR
global N_IN
global N_FC1
global N_OUT
IM_H = IM_W = 80
IM_SIZE = IM_W*IM_H
DIS_FACTOR = 0.99
N_IN = IM_SIZE
N_FC1 = 200
N_OUT = 2
BATCH_SIZE = 10
def prepro(I): # (Game Specific !!!)
""" prepro 210x160x3 uint8 frame into 6400 (80x80) 1D float vector """
I = I[35:195] # crop
I = I[::2,::2,0] # downsample by factor of 2
I[I == 144] = 0 # erase background (background type 1)
I[I == 109] = 0 # erase background (background type 2)
I[I != 0] = 1 # everything else (paddles, ball) just set to 1
return I.astype(np.float).ravel()
def compute_dis_reward(r_list):
dis_reward = 0.
for i in reversed(range(len(r_list))):
dis_reward = DIS_FACTOR*dis_reward + r_list[i]
return dis_reward
def discount_rewards(r):
""" take 1D float array of rewards and compute discounted reward """
discounted_r = np.zeros_like(r)
running_add = 0
for t in reversed(xrange(0, len(r))):
if r[t] != 0: running_add = 0 # reset the sum, since this was a game boundary (pong specific!)
running_add = running_add * DIS_FACTOR + r[t]
discounted_r[t] = running_add
return discounted_r
def sample_prob(y_prob, out_size):
y_acc = []
for idx in range(out_size):
val = y_prob[0, idx] if idx==0 else val+y_prob[0, idx]
y_acc.append(val)
sample_ = tf.random_uniform([1])
dist_list = []
for idx in range(out_size):
dist = y_acc[idx] - sample_
val = tf.select(tf.greater(dist, tf.expand_dims(tf.constant(0.), 0)), dist, tf.expand_dims(tf.constant(1.), 0))
dist_list.append(val)
dist_arr = tf.pack(dist_list)
pos = tf.argmin(dist_arr, 0)
onehot = tf.one_hot(pos, out_size, dtype=tf.float32)
return pos, onehot
def build_fcnet(_x, n_in, n_fc1, n_out):
std1 = np.sqrt(n_in)
std2 = np.sqrt(n_fc1)
with tf.variable_scope("FC1"):
_fc1 = _fc(_x, n_in, n_fc1, std1, activation=True)
with tf.variable_scope("FC2"):
_y = _fc(_fc1, n_fc1, n_out, std2, activation=False)
_y_prob = tf.nn.softmax(_y)
sampled_y, sampled_y_oh = sample_prob(_y_prob, N_OUT)
cost = -tf.reduce_sum(sampled_y_oh*tf.log(_y_prob))
# cost = tf.reduce_sum(tf.nn.softmax_cross_entropy_with_logits(_y))
return cost, sampled_y
def build_agent(_x):
cost, sampled_y = build_fcnet(_x, N_IN, N_FC1, N_OUT)
params = tf.trainable_variables()
grads = tf.gradients(cost, params)
print "grads length: ", len(grads)
print "grads shape: ", grads[0].get_shape()
# global_step = tf.Variable(0, name="global_step", trainable=False)
# optimizer = tf.train.AdamOptimizer(1e-4)
# grads_and_vars = optimizer.compute_gradients(cost)
# train_op = optimizer.apply_gradients(grads_and_vars, global_step=global_step)
# grads_arr = tf.pack(grads)
return sampled_y, grads
def one_episode(env, observation, episode_number):
done = 0
reward = 0
reward_sum = 0
all_gs = []
r_list = []
global prev_x
while done==0:
if render: env.render()
# preprocess the observation, set input to network to be difference image
cur_x = prepro(observation)
x = cur_x - prev_x if prev_x is not None else np.zeros(IM_SIZE)
prev_x = cur_x
feed_dict = {_x: np.expand_dims(x,0)}
fetch = grads
fetch.append(sampled_y)
out_v = sess.run(fetches=fetch, feed_dict=feed_dict)
action_v = out_v[-1]
grads_v = out_v[:-1]
all_gs.append(grads_v)
# action = 2 if np.random.uniform() < action_prob_v else 3 # roll the dice!
action = 2 if action_v==0 else 3 # roll the dice!
observation, reward, done, info = env.step(action)
reward_sum += reward
r_list.append(reward)
if reward != 0: # Pong has either +1 or -1 reward exactly when game ends.
print ('ep %d: game finished, reward: %f' % (episode_number, reward)) + ('' if reward == -1 else ' !!!!!!!!')
# compute the discounted reward backwards through time
dis_reward = discount_rewards(r_list)
# standardize the rewards to be unit normal (helps control the gradient estimator variance)
dis_reward -= np.mean(dis_reward)
dis_reward /= np.std(dis_reward)
game_idx = 0
all_gs_prod = []
for gs in all_gs:
g_prod = []
for g in gs:
g_prod.append(g*dis_reward[game_idx])
all_gs_prod.append(g_prod)
if game_idx == 0:
tot_gs_prod = g_prod
else:
tot_gs_prod = [_p+_q for _p,_q in zip(tot_gs_prod, g_prod)]
game_idx += 1
return tot_gs_prod, reward_sum
_x = tf.placeholder(tf.float32, [None, IM_SIZE])
sampled_y, grads = build_agent(_x)
allvars = tf.trainable_variables()
global_step = tf.Variable(0, name="global_step", trainable=False)
optimizer = tf.train.AdamOptimizer(1e-4)
# optimizer = tf.train.RMSPropOptimizer(1e-4)
init_op = tf.initialize_all_variables()
sess = tf.Session()
sess.run(init_op)
env = gym.make("Pong-v0")
observation = env.reset()
episode_number = 0
batch_idx = 0
running_reward = None
global render
global prev_x
prev_x = None # used in computing the difference frame
render = False
while True:
tot_gs_prod, reward_sum = one_episode(env, observation, episode_number)
# An episode is finished
episode_number += 1
if batch_idx == 0:
batch_grads = tot_gs_prod
else:
for _idx in range(len(batch_grads)):
batch_grads[_idx] += tot_gs_prod[_idx]
if episode_number % BATCH_SIZE == 0:
batch_grads_tf = []
for _idx in range(len(batch_grads)):
batch_grads_tf.append(tf.convert_to_tensor(batch_grads[_idx]))
# Apply batch gradients
print "batch grads length: ", len(batch_grads_tf)
print "batch grads shape: ", type(batch_grads_tf[0]), batch_grads_tf[0].get_shape()
train_op = optimizer.apply_gradients(zip(batch_grads_tf, allvars), global_step=global_step)
if episode_number == BATCH_SIZE:
# uninit_vars = tf.report_uninitialized_variables()
# sess.run(tf.initialize_variables(list(tf.get_variable(name) for name in sess.run(tf.report_uninitialized_variables(tf.all_variables())))))
sess.run(tf.initialize_variables([_v for _v in tf.all_variables() if _v not in allvars]))
sess.run(train_op)
batch_idx = 0
else:
batch_idx += 1
# boring book-keeping
running_reward = reward_sum if running_reward is None else running_reward * 0.99 + reward_sum * 0.01
print 'resetting env. episode reward total was %f. running mean: %f' % (reward_sum, running_reward)
# if episode_number % 100 == 0: pickle.dump(model, open('save.p', 'wb'))
reward_sum = 0
observation = env.reset() # reset env
prev_x = None