-
Notifications
You must be signed in to change notification settings - Fork 4
/
varying_alpha_epsilon_experiment.py
218 lines (187 loc) · 7.83 KB
/
varying_alpha_epsilon_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
import numpy as np
from mdp_matrix import GridWorld, StochasticGridWorld, WindyGridCliffMazeWorld
from double_sarsa import double_sarsa
from expected_sarsa import expected_sarsa
from double_expected_sarsa import double_expected_sarsa
import matplotlib.pyplot as plt
from sarsa import sarsa
# TODO: change these graphs to be over alpha like in the paper
test_rewards = [[i, j, -1.0] for i in range(10) for j in range(10)]
test_rewards[59] = [5, 9, 50]
terminal_states = [59]
obstacles = [[i, j, 0] for i in range(10) for j in range(10)]
obstacles[0*10+3] = [0, 3, 1]
obstacles[0*10+8] = [0, 8, 1]
obstacles[1*10+1] = [1, 1, 1]
obstacles[1*10+4] = [1, 4, 1]
obstacles[1*10+5] = [1, 5, 1]
obstacles[1*10+6] = [1, 6, 1]
obstacles[2*10+1] = [2, 1, 1]
obstacles[2*10+8] = [2, 8, 1]
obstacles[3*10+1] = [3, 1, 1]
obstacles[3*10+4] = [3, 4, 1]
obstacles[3*10+5] = [3, 5, 1]
obstacles[3*10+6] = [3, 6, 1]
obstacles[5*10+0] = [5, 0, 1]
obstacles[5*10+5] = [5, 5, 1]
obstacles[6*10+3] = [6, 3, 1]
obstacles[6*10+4] = [6, 4, 1]
obstacles[6*10+5] = [6, 5, 1]
obstacles[6*10+6] = [6, 6, 1]
obstacles[6*10+8] = [6, 8, 1]
obstacles[7*10+8] = [7, 8, 1]
obstacles[9*10+4] = [9, 4, 1]
traps = [0]*100
traps[4] = 1
traps[9] = 1
traps[13] = 1
traps[33] = 1
traps[43] = 1
traps[47] = 1
traps[67] = 1
traps[72] = 1
traps[96] = 1
start_state = [0, 0]
gw = WindyGridCliffMazeWorld(10, test_rewards, terminal_states, traps, start_state, obstacles)
print test_rewards
average_reward_double_sarsa = []
all_rewards_per_episode_double_sarsa = []
q_var_double_sarsa = []
average_reward_expected_sarsa = []
all_rewards_per_episode_expected_sarsa = []
q_var_expected_sarsa = []
average_reward_double_expected_sarsa = []
all_rewards_per_episode_double_expected_sarsa = []
q_var_double_expected_sarsa = []
average_reward_sarsa = []
all_rewards_per_episode_sarsa = []
q_var_sarsa = []
epsilon = .1
n=1000
alphas = [x for x in np.arange(0.0, 1., .05)]
alphas[0] = .01
# import pdb; pdb.set_trace()
number_of_runs = 5
for r in range(number_of_runs):
for alpha in alphas:
print(alpha)
Q, average_reward, max_reward, all_rewards, Q_variances = double_sarsa(gw, n, epsilon=epsilon, alpha=alpha)
average_reward_double_sarsa.append(average_reward)
all_rewards_per_episode_double_sarsa.append(all_rewards)
q_var_double_sarsa.append(Q_variances)
Q, average_reward, max_reward, all_rewards, Q_variances = expected_sarsa(gw, n, epsilon=epsilon, alpha=alpha)
average_reward_expected_sarsa.append(average_reward)
q_var_expected_sarsa.append(Q_variances)
all_rewards_per_episode_expected_sarsa.append(all_rewards)
Q, average_reward, max_reward, all_rewards, Q_variances = double_expected_sarsa(gw, n, epsilon=epsilon, alpha=alpha)
average_reward_double_expected_sarsa.append(average_reward)
q_var_double_expected_sarsa.append(Q_variances)
all_rewards_per_episode_double_expected_sarsa.append(all_rewards)
Q, average_reward, max_reward, all_rewards, Q_variances = sarsa(gw, n, epsilon=epsilon, alpha=alpha)
q_var_sarsa.append(Q_variances)
average_reward_sarsa.append(average_reward)
all_rewards_per_episode_sarsa.append(all_rewards)
# import pdb; pdb.set_trace()
q_var_sarsa = np.mean(np.mean(np.split(np.array(q_var_sarsa), number_of_runs), axis = 0), axis=1)
q_var_expected_sarsa = np.mean(np.mean(np.split(np.array(q_var_expected_sarsa), number_of_runs), axis = 0), axis=1)
q_var_double_expected_sarsa = np.mean(np.mean(np.split(np.array(q_var_double_expected_sarsa), number_of_runs), axis = 0), axis=1)
q_var_double_sarsa = np.mean(np.mean(np.split(np.array(q_var_double_sarsa), number_of_runs), axis = 0), axis=1)
#
# print("SARSA Mean Q Variance: %f" % np.mean(q_var_sarsa))
# print("Expected SARSA Mean Q Variance: %f" % np.mean(q_var_expected_sarsa))
# print("Double SARSA Mean Q Variance: %f" % np.mean(q_var_double_sarsa))
# print("Double Expected SARSA Mean Q Variance: %f" % np.mean(q_var_double_expected_sarsa))
plt.plot(alphas, q_var_double_sarsa, label="Double Sarsa")
plt.plot(alphas, q_var_expected_sarsa, label="Expected Sarsa")
plt.plot(alphas, q_var_double_expected_sarsa, label="Double Expected Sarsa")
plt.plot(alphas, q_var_sarsa, label="Sarsa")
plt.ylabel('Average Q Variance')
plt.xlabel('alpha')
ax = plt.gca()
# ax.set_xscale('symlog')
ax.legend(loc='upper center', shadow=True)
plt.show()
# TODO: plot all sarsa, expected_sarsa, double_Sarsa
# import pdb; pdb.set_trace()
average_reward_double_sarsa = np.mean(np.split(np.array(average_reward_double_sarsa), number_of_runs), axis=0)
average_reward_expected_sarsa = np.mean(np.split(np.array(average_reward_expected_sarsa), number_of_runs), axis=0)
average_reward_double_expected_sarsa = np.mean(np.split(np.array(average_reward_double_expected_sarsa), number_of_runs), axis=0)
average_reward_sarsa = np.mean(np.split(np.array(average_reward_sarsa), number_of_runs), axis=0)
plt.plot(alphas, average_reward_double_sarsa, label="Double Sarsa")
plt.plot(alphas, average_reward_expected_sarsa, label="Expected Sarsa")
plt.plot(alphas, average_reward_double_expected_sarsa, label="Double Expected Sarsa")
plt.plot(alphas, average_reward_sarsa, label="Sarsa")
plt.ylabel('Average reward')
plt.xlabel('alpha')
ax = plt.gca()
# ax.set_xscale('symlog')
ax.legend(loc='lower center', shadow=True)
plt.show()
# import pdb; pdb.set_trace()
print("Max alpha SARSA: %f" % alphas[np.argmax(average_reward_sarsa)])
print("Max alpha Expected SARSA: %f" % alphas[np.argmax(average_reward_expected_sarsa)])
print("Max alpha Double SARSA: %f" % alphas[np.argmax(average_reward_double_sarsa)])
print("Max alpha Double Expected SARSA: %f" % alphas[np.argmax(average_reward_double_expected_sarsa)])
all_rewards_per_episode_double_sarsa = np.mean(np.split(np.array(all_rewards_per_episode_double_sarsa), number_of_runs), axis=0)
all_rewards_per_episode_expected_sarsa = np.mean(np.split(np.array(all_rewards_per_episode_expected_sarsa), number_of_runs), axis=0)
all_rewards_per_episode_double_expected_sarsa = np.mean(np.split(np.array(all_rewards_per_episode_double_expected_sarsa), number_of_runs), axis=0)
all_rewards_per_episode_sarsa = np.mean(np.split(np.array(all_rewards_per_episode_sarsa), number_of_runs), axis=0)
variance_double_sarsa = np.var(all_rewards_per_episode_double_sarsa, axis=1)
variance_double_expected_sarsa = np.var(all_rewards_per_episode_double_expected_sarsa, axis=1)
variance_expected_sarsa = np.var(all_rewards_per_episode_expected_sarsa, axis=1)
variance_sarsa = np.var(all_rewards_per_episode_sarsa, axis=1)
# import pdb; pdb.set_trace()/
plt.plot(alphas, variance_double_sarsa, label="Double Sarsa")
plt.plot(alphas, variance_expected_sarsa, label="Expected Sarsa")
plt.plot(alphas, variance_double_expected_sarsa, label="Double Expected Sarsa")
plt.plot(alphas, variance_sarsa, label="Sarsa")
plt.ylabel('Variance in Reward')
plt.xlabel('alpha')
ax = plt.gca()
# ax.set_xscale('symlog')
ax.legend(loc='upper center', shadow=True)
plt.show()
#
# for x, e in zip(all_rewards_per_episode_double_sarsa, alphas):
# # import pdb; pdb.set_trace()
# plt.plot(x, label="e=%s"%e)
#
# # break
#
# plt.ylabel('Returns per episode')
# plt.xlabel('episode')
#
# ax = plt.gca()
# # ax.set_xscale('symlog')
# ax.legend(loc='lower right', shadow=True)
# plt.show()
#
#
#
# for x, e in zip(all_rewards_per_episode_expected_sarsa, alphas):
# # import pdb; pdb.set_trace()
# plt.plot(x, label="e=%s"%e)
#
# # break
#
# plt.ylabel('Returns per episode')
# plt.xlabel('episode')
#
# ax = plt.gca()
# # ax.set_xscale('symlog')
# ax.legend(loc='lower right', shadow=True)
# plt.show()
#
# for x, e in zip(all_rewards_per_episode_double_expected_sarsa, alphas):
# # import pdb; pdb.set_trace()
# plt.plot(x, label="e=%s"%e)
#
# # break
#
# plt.ylabel('Returns per episode')
# plt.xlabel('episode')
#
# ax = plt.gca()
# # ax.set_xscale('symlog')
# ax.legend(loc='lower right', shadow=True)
# plt.show()