-
Notifications
You must be signed in to change notification settings - Fork 0
/
RQN_agent.py
321 lines (275 loc) · 14.3 KB
/
RQN_agent.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
#!/usr/bin/env python
# -*- coding: utf-8 -*-
"""
RQN_agent.py
RL training process
Q_agent: RL agent class, contains all networks and training function.
get_q_value()
get_target()
train_network()
get_action()
set_epsilon()
Interaction_env: the environment class that the agent interacts with.
reset()
act()
reward_cal()
action_generation()
"""
import argparse
from interaction_env import Interaction_env
import numpy as np
# import matplotlib.pyplot as plt
import tensorflow as tf
# import keras
# import keras.layers as L
import time
import sys
import json
from config import n_actions, RQN_num_feats, action_length, qlearning_gamma
# n_actions = 6 # 1 no action + 4 directions acc + 1 click
# qlearning_gamma = 0.9
# # n_actions = 4*2 # 4 directions * 2 if click
# action_length = 5 # frames
# RQN_num_feats = 22 # 4 caught object + 2 mouse + 4*4
# Workflow:
# learning_agent.get_action(state_t) -> action ->
# env.step(action) -> (state_next, reward, is_done) ->
# target_agent.get_target(state_next, reward) -> target ->
# learning_agent.train_step(state_t, action, target) -> loss
class Q_agent:
def __init__(self, name, n_actions, qlearning_gamma, input_frames=action_length, num_feats=RQN_num_feats, epsilon=0.99):
self.n_actions = n_actions
self.qlearning_gamma = qlearning_gamma
self.epsilon = epsilon
with tf.variable_scope(name):
self.nn = tf.keras.models.Sequential()
self.nn.add(tf.keras.layers.InputLayer(input_shape=(input_frames,num_feats,)))
# tf.keras.layers.CuDNNLSTM is GPU optimised, switch to tf.keras.layers.LSTM if using CPU
self.nn.add(tf.keras.layers.CuDNNLSTM(22))
# self.nn.add(tf.keras.layers.LSTM(n_state, recurrent_activation='sigmoid'))
self.nn.add(tf.keras.layers.Dense(n_actions))
# Predicting q values for all actions with 5 frames input
self.state_t = tf.placeholder(
tf.float32, [1,input_frames, num_feats])
self.prediction = self.nn(self.state_t)
self.action_t = tf.placeholder(tf.int32)
# self.target_t = tf.placeholder(tf.float32, [1,1])
self.target_t = tf.placeholder(tf.float32)
# mse loss
# self.loss_ = (tf.reduce_sum(self.prediction * tf.one_hot(self.action_t, n_actions), axis=1) - self.target_t) ** 2
self.loss = tf.reduce_mean((tf.reduce_sum(self.prediction * tf.one_hot(self.action_t, n_actions), axis=1) - self.target_t) ** 2)
self.weights = tf.get_collection(
tf.GraphKeys.TRAINABLE_VARIABLES, scope=name)
self.train_step = tf.train.AdamOptimizer(
1e-3).minimize(self.loss, var_list=self.weights)
self.saver = tf.train.Saver(self.weights)
# for forward prediction
def get_q_value(self, state_t, action):
sess = tf.get_default_session()
q_values = sess.run(self.prediction, {self.state_t: state_t})
q_pred_a = tf.reduce_sum(q_values * tf.one_hot(action, n_actions), axis=1)
return q_pred_a
# for target network only
# reward: [batch_size, scalar]
# is_done: [batch_size, bool]
def get_target(self, state_next, reward, is_done):
sess = tf.get_default_session()
q_values_next = sess.run(self.prediction, {self.state_t: state_next})
q_target_a = reward + self.qlearning_gamma * np.amax(q_values_next) # tf.reduce_max(q_values_next, axis=1)
# q_target_a = tf.where(is_done, reward, q_target_a)
return q_target_a
# train network and return loss
# target: calculated from target network
def train_network(self, state_t, action, target):
sess = tf.get_default_session()
_train_step, _loss, _prediction = sess.run([self.train_step, self.loss, self.prediction],
{self.state_t: state_t, self.action_t: action, self.target_t: target})
return _loss
# sample action for a given state
def get_action(self, state_t):
thre = np.random.rand()
if thre < self.epsilon:
action = np.random.choice(n_actions, 1)[0]
else:
sess = tf.get_default_session()
q_values = sess.run(self.prediction, {self.state_t: state_t})
action = np.argmax(q_values)
return action
def set_epsilon(self, epsilon_):
self.epsilon = epsilon_
def load_weigths_into_target_network(agent, target_network):
assigns = []
for w_agent, w_target in zip(agent.weights, target_network.weights):
assigns.append(tf.assign(w_target, w_agent, validate_shape=True))
tf.get_default_session().run(assigns)
# run one episode
# t_max: maximum running time
# train:if True, calculate loss and call train_step
def train_iteration(learning_agent, target_agent, env, t_max, train=False):
session_reward = []
seesion_predictor_loss = []
td_loss = []
s = env.reset() # first 10 frames * 22 num_feats
s=s.reshape((1,action_length,RQN_num_feats))
t = 0
while t < t_max:
a = learning_agent.get_action(s)
# print('action')
# print(a)
trajectory, reward, is_done, predictor_loss = env.act(a)
s_next = trajectory # 10 frames * 22 num_feats
s_next=s_next.reshape((1,action_length,RQN_num_feats))
if train:
target = target_agent.get_target(s_next, reward, is_done)
loss = learning_agent.train_network(s, a, target)
td_loss.append(loss)
session_reward.append(reward)
seesion_predictor_loss.append(predictor_loss)
s = s_next
if is_done:
break
t += action_length
trajectory_history = env.destory()
return session_reward, td_loss, is_done, seesion_predictor_loss, trajectory_history
# Top level training loop, over epochs
def train_loop(learning_agent, target_agent, env, episode, train, timeout, continue_from=0, save_model=False):
# rewards = []
# loss = []
# succeed_episode = 0
# time_taken = []
data = []
for i in range(episode):
# print('[session {} started] '.format(i) + time.strftime("%H:%M:%S", time.localtime()))
session_reward, td_loss, is_done, session_predictor_loss, trajectory_history = train_iteration(learning_agent, target_agent, env, timeout, train)
if not train:
data.append(trajectory_history)
session_reward_mean = np.mean(session_reward)
session_predictor_loss_mean = np.mean(session_predictor_loss)
td_loss_mean = np.mean(td_loss)
print('[session {} finished] '.format(i) + time.strftime("%H:%M:%S", time.localtime()) + ';\t mean reward = {:.4f};\t mean loss = {:.4f};\t total reward = {:.4f};\t epsilon = {:.4f}'.format(
session_reward_mean, td_loss_mean, np.sum(session_reward),learning_agent.epsilon))
print('predictor loss: {}'.format(session_predictor_loss_mean))
# rewards.append(session_reward_mean)
# loss.append(td_loss_mean)
# load_weigths_into_target_network and adjust agent parameters
if train:
if i%2==0:
load_weigths_into_target_network(learning_agent, target_agent)
# learning_agent.set_epsilon(max(learning_agent.epsilon * epsilon_decay, 0.01))
learning_agent.set_epsilon(max(1-i/episode, 0.01))
if i%100==0 and i>0 and save_model:
save_path = learning_agent.saver.save(sess, "./checkpoints/{}_{}_epochs.ckpt".format(exp_name, i + continue_from))
target_save_path = target_agent.saver.save(sess, "./checkpoints/{}_target_{}_epochs.ckpt".format(exp_name, i + continue_from))
print("Model saved in path: %s" % save_path)
# Count and print for catching records
# if is_done:
# succeed_episode += 1
# time_taken.append(len(session_reward))
# print('agent succeed in catching object in {}/{} ({}%) episodes'.format(succeed_episode, episode, succeed_episode/episode*100))
# print('End of training, average actions to catch: {}'.format(np.mean(time_taken)))
if not train:
with open('./model_predictor/data/active_training_data.json', 'w') as data_file:
json.dump(data, data_file, indent=4)
return data
if save_model and train:
# model_json = learning_agent.nn.to_json()
# with open('{}.json'.format(exp_name), 'w') as json_file:
# json_file.write(model_json)
# agent.nn.save_weights('{}.h5'.format(exp_name))
save_path = learning_agent.saver.save(sess, "./checkpoints/{}_{}_epochs.ckpt".format(exp_name, episode + continue_from))
target_save_path = target_agent.saver.save(sess, "./checkpoints/{}_target_{}_epochs.ckpt".format(exp_name, episode + continue_from))
print("Model saved in path: %s" % save_path)
print("Model saved!")
# np.savetxt('{}.txt'.format(exp_name), (rewards, loss))
# print("Training details saved!")
return None
if __name__ == "__main__":
parser = argparse.ArgumentParser(
description='training a recurrent q-network')
parser.add_argument('--episode', type=int, action='store',
help='number of epoches to train', default=50)
parser.add_argument('--save_model', type=bool, action='store',
help='save trained model or not', default=False)
parser.add_argument('--active_learning', type=bool, action='store',
help='load pretrained RQN & model predictor to do active learning', default=False)
parser.add_argument('--train', type=bool, action='store',
help='if to train a model', default=False)
parser.add_argument('--epsilon', type=float, action='store',
help='epsilon for Q learning', default=0.99)
parser.add_argument('--lr', type=float, action='store',
help='learning rate for Adam optimiser', default=1e-4)
parser.add_argument('--timeout', type=int, action='store',
help='max number of frames for one episode, 1/60s per frame', default=1800)
parser.add_argument('--continue_from', type=int, action='store',
help='continue training from previous trained model', default=0)
args = parser.parse_args()
# RQN_num_feats = 22
# input_frames = 5
epsilon_decay = 0.9
if args.episode >= 10000:
epsilon_decay = 0.9995 # 10000 epochs
elif args.episode >= 1000:
epsilon_decay = 0.999 # 2000 epochs
else:
epsilon_decay = 0.95
# exp_name = 'RQN_20_{:1.0e}'.format(args.lr) # 20 reward for successful catching + bounded getting close reward
# exp_name = 'RQN_bonded_{:1.0e}'.format(args.lr) # reward for getting close to nearest puck is bounded
# exp_name = 'RQN_more_reward_{:1.0e}'.format(args.lr) # add reward for getting close to nearest puck
# exp_name = 'RQN_{:1.0e}'.format(args.lr) # only 5 reward for successful catching
exp_name = 'new_active_learning_world-1'
# exp_name = 'new_catch_training'
# tf.reset_default_graph()
# sess = tf.InteractiveSession()
# keras.backend.set_session(sess)
# initialize interaction_env
environment = Interaction_env()
# initialize learning_agent and target_agent
if not args.active_learning:
# train for catching pucks
if args.train:
rqn_agent_graph = tf.Graph()
with rqn_agent_graph.as_default():
learning_agent = Q_agent("learning_agent", n_actions, qlearning_gamma, epsilon=args.epsilon)
target_agent = Q_agent("target_agent", n_actions, qlearning_gamma, epsilon=args.epsilon)
sess = tf.InteractiveSession(graph = rqn_agent_graph)
sess.run(tf.global_variables_initializer())
# environment.predictor.saver.restore(sess, "./model_predictor/checkpoints/pretrained_model_predictor_2.ckpt")
if args.continue_from > 0:
learning_agent.saver.restore(sess, "./checkpoints/{}_{}_epochs.ckpt".format(exp_name,args.continue_from))
target_agent.saver.restore(sess, "./checkpoints/{}_target_{}_epochs.ckpt".format(exp_name,args.continue_from))
else:
if args.continue_from == 0:
sys.exit('[ERROR] test model not specified')
rqn_agent_graph = tf.Graph()
with rqn_agent_graph.as_default():
learning_agent = Q_agent("learning_agent", n_actions, qlearning_gamma, epsilon=0)
sess = tf.InteractiveSession(graph = rqn_agent_graph)
sess.run(tf.global_variables_initializer())
learning_agent.saver.restore(sess, "./checkpoints/{}_{}_epochs.ckpt".format(exp_name,args.continue_from))
target_agent = None
else:
# train for active learning
if args.train:
rqn_agent_graph = tf.Graph()
with rqn_agent_graph.as_default():
learning_agent = Q_agent("learning_agent", n_actions, qlearning_gamma, epsilon=args.epsilon)
target_agent = Q_agent("target_agent", n_actions, qlearning_gamma, epsilon=args.epsilon)
sess = tf.InteractiveSession(graph = rqn_agent_graph)
sess.run(tf.global_variables_initializer())
# RQN agent trained on 10000 episodes with bonded reward "./checkpoints/RQN_bonded_1e-04_10000_epochs.ckpt"
learning_agent.saver.restore(sess, "./checkpoints/trained_RQN_catching.ckpt")
target_agent.saver.restore(sess, "./checkpoints/trained_RQN_catching_target.ckpt")
else:
print('Active data generation started...')
if args.continue_from == 0:
sys.exit('[ERROR] test model not specified')
rqn_agent_graph = tf.Graph()
with rqn_agent_graph.as_default():
learning_agent = Q_agent("learning_agent", n_actions, qlearning_gamma, epsilon=0)
sess = tf.InteractiveSession(graph = rqn_agent_graph)
sess.run(tf.global_variables_initializer())
# RQN agent trained on 10000 episodes with bonded reward "./checkpoints/RQN_bonded_1e-04_10000_epochs.ckpt"
learning_agent.saver.restore(sess, "./checkpoints/active_learning_loss_reward_world-1_{}_epochs.ckpt".format(args.continue_from))
target_agent = None
# train
_ = train_loop(learning_agent, target_agent, environment, args.episode, args.train, args.timeout, args.continue_from, args.save_model)