This repository has been archived by the owner on May 30, 2024. It is now read-only.
-
Notifications
You must be signed in to change notification settings - Fork 0
/
spcrl_grid_discrete.py
241 lines (210 loc) · 8.56 KB
/
spcrl_grid_discrete.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
# A Performance-Based Start State Curriculum Framework for Reinforcement Learning
# Copyright (c) 2021 Robert Bosch GmbH
#
# This program is free software: you can redistribute it and/or modify
# it under the terms of the GNU Affero General Public License as published
# by the Free Software Foundation, either version 3 of the License, or
# (at your option) any later version.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU Affero General Public License for more details.
#
# You should have received a copy of the GNU Affero General Public License
# along with this program. If not, see <https://www.gnu.org/licenses/>.
# IMPORT #
import os
import argparse
import copy
import random
import time
import numpy as np
import torch
from lib.trpo.trpo_discrete_aug_33 import TRPOTrainer
from lib.start_selection.reaching_probability_map import generate_reach_prob_map_oracle
from lib.start_selection.select_start import start_state_sampling_probabilities_from_rp
def run_fun(seed_list,
grid_var,
num_inner_episodes,
num_outer_steps,
policy_hidden_sizes_list,
max_kl_val,
damp_val,
max_iter_val,
batch_size_val,
seed_arg,
lr_arg,
cluster=False,
num_eval=10
):
# INITIALIZATION
start_time = time.time()
prefix = 'spcrl_grid_discrete'
suffix = '%s_%i_%s' % (lr_arg, grid_var, seed_arg)
config_dict = {
'prefix': prefix,
'suffix': suffix,
'barcode': int(start_time),
'seed_list': seed_list,
'grid_var': grid_var,
'num_inner_episodes': num_inner_episodes,
'num_outer_steps': num_outer_steps,
'policy_hidden_sizes_list': policy_hidden_sizes_list,
'max_kl_val': max_kl_val,
'damp_val': damp_val,
'max_iter_val': max_iter_val,
'batch_size_val': batch_size_val,
}
if cluster:
save_path = os.environ.get('SSD') + "/" + os.environ.get('EXP') + "/"
else:
save_path = ""
f = open(save_path + "config_" + prefix + "_" + str(config_dict['barcode']) + "_" + suffix + '.txt', 'w')
f.write(str(config_dict))
f.close()
print("CONFIG")
print(config_dict)
# LOOP OVER SEEDS #
for seed_val in seed_list:
# SEED #
torch.manual_seed(seed_val)
random.seed(seed_val)
np.random.seed(seed_val)
# INITIALIZATION #
myTRPO = TRPOTrainer(grid_type=grid_var,
state_dim=10,
action_dim=4,
hidden_sizes=policy_hidden_sizes_list,
max_kl=max_kl_val,
damping=damp_val,
batch_size=batch_size_val,
inner_episodes=num_inner_episodes,
max_iter=max_iter_val
)
eps_tot = 0
acc_eps_list = [0]
acc_result_list_ust = [0]
start_save = []
rpm_save = []
print("\n")
print("SEED")
print(seed_val)
print("\n")
i = 0
while eps_tot < num_outer_steps:
i += 1
print(eps_tot)
rpm = generate_reach_prob_map_oracle(trpo=myTRPO, state_list=myTRPO.default_starts)
rpm2save = copy.deepcopy(rpm)
rpm_save.append(rpm2save)
# print(rpm)
train_starts, sampling_prob_list = start_state_sampling_probabilities_from_rp(
trpo=myTRPO,
start_candidates=myTRPO.default_starts,
rpm=rpm,
step=eps_tot,
temp_param_mode='exp')
print("TRAIN STARTS")
# print(train_starts)
# print(sampling_prob_list)
print(not train_starts)
if not train_starts:
_, result, s_list, a_list, _, st_list, _ = myTRPO.train(start_p_list=[],
goal_p=myTRPO.grid.goal,
sampling_probs=[],
sampling_mode=False)
else:
_, result, s_list, a_list, _, st_list, _ = myTRPO.train(start_p_list=train_starts,
goal_p=myTRPO.grid.goal,
sampling_probs=sampling_prob_list,
sampling_mode=True)
start_save.append(st_list)
print("RESULT")
print(result)
eps_tot += num_inner_episodes
print("TEST UNIFORM")
test_starts = myTRPO.grid.sample_random_pos(number=num_eval)
print("TEST STARTS")
print(test_starts)
reach_prob_ust = 0
for start_state in test_starts:
_, result_ust, _, _, _ = myTRPO.test(start_p_list=start_state, goal_p=myTRPO.grid.goal)
reach_prob_ust += result_ust
# print(result_ust)
reach_prob_ust = reach_prob_ust / len(test_starts)
print("Iteration: %i" % i)
print("Reach Prob UST: %.2f" % reach_prob_ust)
print("Time passed (in min): %.2f" % ((time.time() - start_time) / 60))
acc_eps_list.append(eps_tot)
acc_result_list_ust.append(reach_prob_ust)
reach_prob_list_ust = acc_result_list_ust
# SAVE RESULTS #
episodes_np = np.array(acc_eps_list)
np.save(save_path + "episodes_" + prefix + "_" + str(seed_val) + "_" + str(grid_var) + "_" + suffix + "_" +
str(config_dict['barcode']), episodes_np)
results_ust_np = np.array(reach_prob_list_ust)
np.save(save_path + "results_ust_" + prefix + "_" + str(seed_val) + "_" + str(grid_var) + "_" + suffix + "_" +
str(config_dict['barcode']), results_ust_np)
st_np = np.array(start_save)
np.save(save_path + "st_" + prefix + "_" + str(seed_val) + "_" + str(grid_var) + "_" + suffix + "_" +
str(config_dict['barcode']), st_np)
rpm_np = np.array(rpm_save)
np.save(save_path + "rpm_" + prefix + "_" + str(seed_val) + "_" + str(grid_var) + "_" + suffix + "_" +
str(config_dict['barcode']), rpm_np)
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument("--seed", type=str, default="s1")
parser.add_argument("--cluster", type=bool, default=False)
args = parser.parse_args()
if args.seed == "s1":
seed_l = [1535719580]
elif args.seed == "s2":
seed_l = [1535720536]
elif args.seed == "s3":
seed_l = [1535721129]
elif args.seed == "s4":
seed_l = [1535721985]
elif args.seed == "s5":
seed_l = [1535723522]
elif args.seed == "s6":
seed_l = [1535724275]
elif args.seed == "s7":
seed_l = [1535726291]
elif args.seed == "s8":
seed_l = [1535954757]
elif args.seed == "s9":
seed_l = [1535957367]
elif args.seed == "s10":
seed_l = [1535953242]
elif args.seed == "ps1":
seed_l = [1535719580, 1535720536, 1535721129]
elif args.seed == "ps2":
seed_l = [1535721985, 1535723522, 1535724275]
elif args.seed == "ps3":
seed_l = [1535726291, 1535954757]
elif args.seed == "ps4":
seed_l = [1535957367, 1535953242]
elif args.seed == "p1":
seed_l = [1535719580, 1535720536, 1535721129, 1535721985, 1535723522]
elif args.seed == "p2":
seed_l = [1535724275, 1535726291, 1535954757, 1535957367, 1535953242]
elif args.seed == "all":
seed_l = [1535719580, 1535720536, 1535721129, 1535721985, 1535723522,
1535724275, 1535726291, 1535954757, 1535957367, 1535953242]
else:
seed_l = [0]
run_fun(seed_list=seed_l,
grid_var=6,
num_inner_episodes=5,
num_outer_steps=5000,
policy_hidden_sizes_list=[64, 64, 64],
max_kl_val=5e-4,
damp_val=5e-3,
max_iter_val=100,
batch_size_val=3200,
seed_arg=args.seed,
lr_arg="llr",
cluster=args.cluster
)
# 1535719580, 1535720536, 1535721129, 1535721985, 1535723522, 1535724275, 1535726291, 1535954757, 1535957367, 1535953242