-
Notifications
You must be signed in to change notification settings - Fork 0
/
experiment.py
238 lines (168 loc) · 8.05 KB
/
experiment.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
from utils.utils import plot_graphs
import numpy as np
class Experiment(object):
def __init__(self, env, agent):
self.env = env
self.agent = agent
self.episode_length = np.array([0])
self.episode_reward = np.array([0])
def run(self, max_number_of_episodes=100,
interactive=False, display_frequency=1):
# repeat for each episode
for episode_number in range(max_number_of_episodes):
# no need to initialize state
self.env.reset()
done = False # used to indicate terminal state
R = 0 # used to display accumulated rewards for an episode
t = 0 # used to display accumulated steps for an episode
# repeat for each step of episode, until state is terminal
while not done:
# increase step counter - for display
t += 1
# choose action from state
action = self.agent.act()
# take action, observe reward and next state
next_state, reward, done, _ = self.env.step(action)
# accumulate reward - for display
R += reward
# if interactive display, show update for each step
if interactive:
self.env.render()
# keep episode length - for display
self.episode_length = np.append(self.episode_length, t)
# keep episode reward - for display
self.episode_reward = np.append(self.episode_reward, R)
# if interactive display, show update for the episode
if interactive:
self.env.close()
plot_graphs(self.episode_reward, self.episode_length)
def run_randomsearch(self, max_number_of_episodes=100, interactive=False,
display_frequency=1):
# repeat for each episode
for episode in range(max_number_of_episodes):
state = self.env.reset() # Initialization
done = False # used to indicate terminal state
R = 0 # used to display accumulated rewards for an episode
t = 0 # used to display accumulated steps for an episode
# repeat for each step of episode, until state is terminal
while not done:
t += 1 # increase step counter - for display
# choose action from state using policy derived from Q
action = self.agent.act(state)
# take action, observe reward and next state
state, reward, done, _ = self.env.step(action)
R += reward # accumulate reward - for display
# if interactive display, show update for each step
if interactive:
self.env.render()
self.agent.learn(R, done)
self.episode_length = np.append(
self.episode_length, t) # keep episode length - for display
self.episode_reward = np.append(
self.episode_reward, R) # keep episode reward - for display
# if interactive display, show update for the episode
if interactive:
self.env.render()
self.env.close()
def run_qlearning(self, max_number_of_episodes=100,
interactive=False, display_frequency=1, debug=True):
# repeat for each episode
for episode_number in range(max_number_of_episodes):
# initialize state
state = self.env.reset()
done = False # used to indicate terminal state
R = 0 # used to display accumulated rewards for an episode
t = 0 # used to display accumulated steps for an episode
# repeat for each step of episode, until state is terminal
while not done:
# choose action from state using policy derived from Q
action = self.agent.act(state)
# increment the lenght of the episode
t += 1
# take action, observe reward and next state
next_state, reward, done, _ = self.env.step(action)
# agent learn
self.agent.learn(state, action, reward, next_state)
# update state & action
state = next_state
R += reward # accumulate reward - for display
# if interactive display, show update for each step
if interactive:
self.env.render()
# sleep(0.1)
if debug:
print("\nEpisode = %d" % episode_number)
print("t = %d" % t)
print("Action: %d" % action)
print("State: %s" % str(state))
print("Reward: %f" % reward)
# keep episode length - for display
self.episode_length = np.append(self.episode_length, t)
# keep episode reward - for display
self.episode_reward = np.append(self.episode_reward, R)
# update exploration rate
self.agent.epsilon = self.agent.get_explore_rate(episode_number)
# update learning rate
self.agent.alpha = self.agent.get_learning_rate(episode_number)
# if interactive display, show update for the episode
if interactive:
self.env.close()
plot_graphs(self.episode_reward, self.episode_length)
def run_montecarlo(self, max_number_of_episodes=100,
interactive=False, debug=False):
# repeat for each episode
for episode_number in range(max_number_of_episodes):
# no need to initialize state
state = self.env.reset()
# keep state tracking
states = []
# keep action tracking
actions = []
# keep track of rewards
rewards = []
done = False # used to indicate terminal state
R = 0 # used to display accumulated rewards for an episode
t = 0 # used to display accumulated steps for an episode
# repeat for each step of episode, until state is terminal
while not done:
# track states
states.append(state)
# increase step counter - for display
t += 1
# choose action from state
action = self.agent.act(state)
# update actions
actions.append(action)
# take action, observe reward and next state
next_state, reward, done, _ = self.env.step(action)
# update state
state = next_state
# accumulate reward - for display
R += reward
rewards.append(reward)
# if interactive display, show update for each step
if interactive:
self.env.render()
# print evolution
if debug:
print("\nEpisode = %d" % episode_number)
print("t = %d" % t)
print("Action: %d" % action)
print("State: %s" % str(state))
print("State Approximation: %s" % str(self.agent.state_to_bucket(state)))
print("Reward: %f" % reward)
# learn
self.agent.learn(states, actions, rewards)
# keep episode length - for display
self.episode_length = np.append(self.episode_length, t)
# keep episode reward - for display
self.episode_reward = np.append(self.episode_reward, R)
# update exploration rate
self.agent.epsilon = self.agent.get_explore_rate(episode_number)
# update learning rate
self.agent.alpha = self.agent.get_learning_rate(episode_number)
print(self.agent.epsilon, self.agent.alpha, R)
# if interactive display, show update for the episode
if interactive:
self.env.close()
plot_graphs(self.episode_reward, self.episode_length)