-
Notifications
You must be signed in to change notification settings - Fork 0
/
Q_leaning.py
153 lines (117 loc) · 5.17 KB
/
Q_leaning.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
import tensorflow as tf
import numpy as np
import gym
import seaborn as sns
import matplotlib.pyplot as plt
import time
start = time.clock()
class DQN(object):
def __init__(self, n_a, s_dim, r_q = 1e-3, gamma = 0.9, tau = 0.01,
batch_size =32, memory_capacity = 10000, double_q = False):
self.a_dim = 1
self.memory = np.zeros((memory_capacity, s_dim * 2 + self.a_dim + 2), dtype=np.float32)
self.memory_size = memory_capacity
self.pointer = 0
self.batch_size = batch_size
self.lr_q, self.gamma, self.tau = r_q, gamma, tau
self.n_a, self.s_dim = n_a, s_dim
self.s = tf.placeholder(tf.float32, [None, s_dim], 's')
self.a = tf.placeholder(tf.int32, [None], 'a')
self.s_next = tf.placeholder(tf.float32, [None, s_dim], 's_')
self.r = tf.placeholder(tf.float32, [None, 1], 'r')
self.done = tf.placeholder(tf.float32, [None, 1], 'done')
self.double_q = double_q
with tf.variable_scope('eval'):
self.q = self.build_DQN(self.s)
with tf.variable_scope('target'):
q_ = self.build_DQN(self.s_next, False)
self.sample_a = tf.argmax(self.q, axis=1)
# if self.double_q:
# with tf.variable_scope('eval', reuse=True):
# q_reuse = self.build_DQN(self.s_next)
# max_a = tf.argmax(q_reuse, axis=1)
# one_hot_mask = tf.one_hot(max_a, depth=self.n_a)
# q_target = self.r + gamma * tf.reduce_sum(q_ * one_hot_mask, axis=1, keepdims=True) * (1.0 - self.done)
# else:
q_target = self.r + gamma * tf.reduce_max(q_, axis=1) * (1.0 - self.done)
q_eval = tf.reduce_sum(self.q * tf.one_hot(self.a, depth = self.n_a), axis=1, keepdims=True)
# in the feed_dic for the td_error, the self.a should change to actions in memory
self.td_error = tf.losses.mean_squared_error(labels=q_target, predictions=q_eval)
# networks parameters
self.eval_para = tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES, scope='eval')
self.target_para = tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES, scope='target')
self.train_op_q = tf.train.AdamOptimizer(self.lr_q).minimize(self.td_error, var_list=self.eval_para)
self.soft_replace = [tf.assign(target, var) for target, var in zip(self.target_para, self.eval_para)]
self.sess = tf.Session()
self.sess.run(tf.global_variables_initializer())
def sample_action(self, s):
a = self.sess.run(self.sample_a, feed_dict={self.s: s[None, :]})[0]
return a
def learn(self):
# soft target replacement
self.sess.run(self.soft_replace)
indices = np.random.choice(self.memory_size, size=self.batch_size)
bt = self.memory[indices, :]
s = bt[:, :self.s_dim]
a = bt[:, self.s_dim: self.s_dim + self.a_dim]
r = bt[:, -self.s_dim - 1: -self.s_dim]
s_next = bt[:, -self.s_dim-1:-1]
done = bt[:, [-1]]
_ , lose = self.sess.run([self.train_op_q, self.td_error], {self.s: s, self.a: a.ravel(),
self.r: r, self.s_next: s_next,
self.done: done})
return lose
def store_transition(self, s, a, r, s_, done):
transition = np.hstack((s, [a], [r], s_, [done]))
index = self.pointer % self.memory_size # replace the old memory with new memory
self.memory[index, :] = transition
self.pointer += 1
def build_DQN(self, s, trainable = True):
hidden_units = [20, 10]
activation = [tf.nn.relu, tf.nn.relu]
layer = s
for units, activ in zip(hidden_units, activation):
layer = tf.layers.dense(layer, units, activation=activ, trainable=trainable)
q = tf.layers.dense(layer, self.n_a, activation=None, trainable=trainable)
return q
if __name__ == '__main__':
MAX_EPISODES = 500
MAX_EP_STEPS = 200
RENDER = False
ENV_NAME = 'Boxing-ram-v0'
env = gym.make(ENV_NAME)
env = env.unwrapped
env.seed(1)
s_dim = env.observation_space.shape[0]
a_dim = env.action_space.n
dqn = DQN(a_dim, s_dim)
episode_reward = []
for i in range(MAX_EPISODES):
s = env.reset()
ep_reward = 0
for j in range(MAX_EP_STEPS):
if RENDER:
env.render()
# Add exploration noise
a = dqn.sample_action(s)
s_, r, done, info = env.step(a)
dqn.store_transition(s, a, r, s_, done)
if dqn.pointer > dqn.memory_size:
loss = dqn.learn()
print('loss at epoch %s is %s' %(i, loss))
s = s_
ep_reward += r
if done: break
episode_reward.append(ep_reward)
print('Episode:', i, ' Reward: %i' % int(ep_reward))
if ep_reward > 10:
RENDER = True
end = time.clock()
print('Running time: %s Seconds' % (end - start))
sns.set(style="darkgrid")
plt.figure(1)
plt.plot(episode_reward, label='DDPG')
plt.xlabel('Episode')
plt.ylabel('Reward')
plt.legend(loc='best')
plt.show()