-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathconfig.py
238 lines (217 loc) · 12.9 KB
/
config.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
import argparse
import random
from time import time
import numpy as np
import torch
def default_config():
return dict(
seed_runs=1000,
seed_start=0,
dt=0.1,
dt_simulation=0.01,
mppi_roll_outs=1000,
mppi_time_steps=20,
mppi_lambda=0.01,
mppi_sigma=1.0,
observation_noise=0.01,
observing_cost=50,
observing_fixed_frequency=1,
discrete_planning=True,
discrete_interval=10,
continuous_time_threshold=1.0, # From [0,1]
# ==============Expert Dataset collection
collect_expert_samples=1e6,
collect_expert_force_generate_new_data=False,
collect_expert_random_action_noise=1.0,
collect_expert_cores_per_env_sampler=16,
collect_expert_episodes_per_sampler_task=1,
train_with_expert_trajectories=False,
offline_datasets_path="./offlinedata/",
# ==============Model parameters
saved_models_path="./saved_models/",
normalize=True,
normalize_time=True,
model_pe_hidden_units=256,
encode_obs_time=False,
model_seed=0,
# ====New parameters
model_ensemble_size=5,
model_pe_activation="tanh",
model_pe_initialization="xavier",
model_pe_use_pets_log_var=True,
# ==============Training parameters
weight_decay=0,
learning_rate=1e-4,
training_epochs=10000000,
training_batch_size=16,
iters_per_log=500,
clip_grad_norm=0.1,
clip_grad_norm_on=False,
train_dt_multiple=1,
ts_grid="exp", # ['fixed', 'uniform', 'exp']
train_samples_per_dim=10,
iters_per_evaluation=1e15,
lr_scheduler_step_size=20,
lr_scheduler_gamma=0.1,
use_lr_scheduler=False,
reuse_state_actions_when_sampling_times=False,
end_training_after_seconds=int(1350 * 6.0),
rand_sample=True,
# ==============Misc
log_folder="logs",
save_video=False,
plot_telem=False,
sweep_mode=False,
torch_deterministic=True,
multi_process_results=True,
retrain=False,
force_retrain=False,
start_from_checkpoint=True,
print_settings=False,
training_use_only_samples=None,
friction=False,
wandb_project="ActiveObservingControl",
oracle_var_type="state_oracle_var",
special_mode_continuous_planning_execute_only_n_actions=None,
use_95_ci=True,
)
def parse_args(config):
# fmt: off
parser = argparse.ArgumentParser()
parser.add_argument("--seed_runs", type=int, default=config['seed_runs'], help="seed_runs")
parser.add_argument("--retrain", choices=('True','False'), default=str(config['retrain']), help="retrain")
parser.add_argument("--force_retrain", choices=('True','False'), default=str(config['force_retrain']), help="force_retrain")
parser.add_argument("--start_from_checkpoint", choices=('True','False'), default=str(config['start_from_checkpoint']), help="start_from_checkpoint")
parser.add_argument("--print_settings", choices=('True','False'), default=str(config['print_settings']), help="print_settings")
parser.add_argument("--seed_start", type=int, default=config['seed_start'], help="seed_start")
parser.add_argument("--dt", type=float, default=config['dt'], help="dt")
parser.add_argument("--learning_rate", type=float, default=config['learning_rate'], help="learning_rate")
parser.add_argument("--collect_expert_samples", type=float, default=config['collect_expert_samples'], help="collect_expert_samples")
parser.add_argument("--training_epochs", type=int, default=config['training_epochs'], help="training_epochs")
parser.add_argument("--training_batch_size", type=int, default=config['training_batch_size'], help="training_batch_size")
parser.add_argument("--saved_models_path", type=str, default=config['saved_models_path'], help="saved_models_path")
parser.add_argument("--offline_datasets_path", type=str, default=config['offline_datasets_path'], help="offline_datasets_path")
parser.add_argument("--iters_per_log", type=int, default=config['iters_per_log'], help="iters_per_log")
parser.add_argument("--clip_grad_norm", type=float, default=config['clip_grad_norm'], help="clip_grad_norm")
parser.add_argument("--clip_grad_norm_on", choices=('True','False'), default=str(config['clip_grad_norm_on']), help="clip_grad_norm_on")
parser.add_argument("--collect_expert_cores_per_env_sampler", type=float, default=config['collect_expert_cores_per_env_sampler'], help="collect_expert_cores_per_env_sampler")
parser.add_argument("--collect_expert_episodes_per_sampler_task", type=float, default=config['collect_expert_episodes_per_sampler_task'], help="collect_expert_episodes_per_sampler_task")
parser.add_argument("--normalize", choices=('True','False'), default=str(config['normalize']), help="normalize")
parser.add_argument("--normalize_time", choices=('True','False'), default=str(config['normalize_time']), help="normalize_time")
parser.add_argument("--train_dt_multiple", type=float, default=config['train_dt_multiple'], help="train_dt_multiple")
parser.add_argument("--collect_expert_random_action_noise", type=float, default=config['collect_expert_random_action_noise'], help="collect_expert_random_action_noise")
parser.add_argument("--ts_grid", type=str, default=config['ts_grid'], help="ts_grid")
parser.add_argument("--train_samples_per_dim", type=int, default=config['train_samples_per_dim'], help="train_samples_per_dim")
parser.add_argument("--model_pe_hidden_units", type=int, default=config['model_pe_hidden_units'], help="model_pe_hidden_units")
parser.add_argument("--lr_scheduler_step_size", type=int, default=config['lr_scheduler_step_size'], help="lr_scheduler_step_size")
parser.add_argument("--lr_scheduler_gamma", type=float, default=config['lr_scheduler_gamma'], help="lr_scheduler_gamma")
parser.add_argument("--weight_decay", type=float, default=config['weight_decay'], help="weight_decay")
parser.add_argument("--log_folder", type=str, default=config['log_folder'], help="log_folder")
parser.add_argument("--iters_per_evaluation", type=float, default=config['iters_per_evaluation'], help="iters_per_evaluation")
parser.add_argument("--mppi_roll_outs", type=int, default=config['mppi_roll_outs'], help="mppi_roll_outs")
parser.add_argument("--mppi_time_steps", type=int, default=config['mppi_time_steps'], help="mppi_time_steps")
parser.add_argument("--mppi_lambda", type=float, default=config['mppi_lambda'], help="mppi_lambda")
parser.add_argument("--mppi_sigma", type=float, default=config['mppi_sigma'], help="mppi_sigma")
parser.add_argument("--encode_obs_time", choices=('True','False'), default=str(config['encode_obs_time']), help="encode_obs_time")
parser.add_argument("--reuse_state_actions_when_sampling_times", choices=('True','False'), default=str(config['reuse_state_actions_when_sampling_times']), help="reuse_state_actions_when_sampling_times")
parser.add_argument("--model_seed", type=int, default=config['model_seed'], help="model_seed")
parser.add_argument("--save_video", choices=('True','False'), default=str(config['save_video']), help="save_video")
parser.add_argument("--sweep_mode", choices=('True','False'), default=str(config['sweep_mode']), help="sweep_mode")
parser.add_argument("--rand_sample", choices=('True','False'), default=str(config['rand_sample']), help="rand_sample")
parser.add_argument("--collect_expert_force_generate_new_data", choices=('True','False'), default=str(config['collect_expert_force_generate_new_data']), help="collect_expert_force_generate_new_data")
parser.add_argument("--train_with_expert_trajectories", choices=('True','False'), default=str(config['train_with_expert_trajectories']), help="train_with_expert_trajectories")
parser.add_argument("--end_training_after_seconds", type=float, default=config['end_training_after_seconds'], help="end_training_after_seconds")
parser.add_argument("--torch_deterministic", choices=('True','False'), default=str(config['torch_deterministic']), help="torch_deterministic")
parser.add_argument("--use_lr_scheduler", choices=('True','False'), default=str(config['use_lr_scheduler']), help="use_lr_scheduler")
parser.add_argument("--multi_process_results", choices=('True','False'), default=str(config['multi_process_results']), help="multi_process_results")
parser.add_argument("--observation_noise", type=float, default=config['observation_noise'], help="observation_noise")
parser.add_argument("--friction", choices=('True','False'), default=str(config['friction']), help="friction")
args = parser.parse_args()
ddict = vars(args)
ddict['normalize'] = ddict['normalize'] == 'True'
ddict['normalize_time'] = ddict['normalize_time'] == 'True'
ddict['encode_obs_time'] = ddict['encode_obs_time'] == 'True'
ddict['reuse_state_actions_when_sampling_times'] = ddict['reuse_state_actions_when_sampling_times'] == 'True'
ddict['save_video'] = ddict['save_video'] == 'True'
ddict['sweep_mode'] = ddict['sweep_mode'] == 'True'
ddict['rand_sample'] = ddict['rand_sample'] == 'True'
ddict['collect_expert_force_generate_new_data'] = ddict['collect_expert_force_generate_new_data'] == 'True'
ddict['train_with_expert_trajectories'] = ddict['train_with_expert_trajectories'] == 'True'
ddict['torch_deterministic'] = ddict['torch_deterministic'] == 'True'
ddict['use_lr_scheduler'] = ddict['use_lr_scheduler'] == 'True'
ddict['multi_process_results'] = ddict['multi_process_results'] == 'True'
ddict['retrain'] = ddict['retrain'] == 'True'
ddict['force_retrain'] = ddict['force_retrain'] == 'True'
ddict['start_from_checkpoint'] = ddict['start_from_checkpoint'] == 'True'
ddict['print_settings'] = ddict['print_settings'] == 'True'
ddict['friction'] = ddict['friction'] == 'True'
ddict['clip_grad_norm_on'] = ddict['clip_grad_norm_on'] == 'True'
return ddict
def get_config():
defaults = default_config()
args = parse_args(defaults)
defaults.update(args)
return defaults
class dotdict(dict):
"""dot.notation access to dictionary attributes"""
__getattr__ = dict.get
__setattr__ = dict.__setitem__
__delattr__ = dict.__delitem__
def default_config_dd():
d_c = default_config()
return dotdict(d_c)
def seed_all(seed=None):
"""
Set the torch, numpy, and random module seeds based on the seed
specified in config. If there is no seed or it is None, a time-based
seed is used instead and is written to config.
"""
# Default uses current time in milliseconds, modulo 1e9
if seed is None:
seed = round(time() * 1000) % int(1e9)
# Set the seeds using the shifted seed
torch.manual_seed(seed)
np.random.seed(seed)
random.seed(seed)
def load_model_cuda_memory_details():
return {
"largest_model_pe": 1272, # MiB
"largest_model_oracle": 2794, #
}
def load_observing_var_thresholds():
return {
0.05: {
"oderl-cartpole": {"continuous": 0.025, "discrete": 0.25},
"oderl-pendulum": {"continuous": 0.025, "discrete": 0.25},
"oderl-acrobot": {"continuous": 0.025, "discrete": 0.25},
"oderl-cancer": {"continuous": 1.0, "discrete": 1.5},
},
0.1: {
"oderl-cartpole": {"continuous": 0.029934801, "discrete": 0.5},
"oderl-pendulum": {"continuous": 0.012269268, "discrete": 0.061973028},
"oderl-acrobot": {"continuous": 0.08927406, "discrete": 0.28180087},
"oderl-cancer": {"continuous": 2.9376497, "discrete": 2.5688453},
},
0.2: {
"oderl-cartpole": {"continuous": 0.06576242, "discrete": 0.8842558},
"oderl-pendulum": {"continuous": 0.04570341, "discrete": 0.4898505},
"oderl-acrobot": {"continuous": 0.27656594, "discrete": 1.6966783},
"oderl-cancer": {"continuous": 3.657069, "discrete": 4.8863864},
},
0.4: {
"oderl-cartpole": {"continuous": 0.19495021, "discrete": 72.789734},
"oderl-pendulum": {"continuous": 0.046161246, "discrete": 1.2940274},
"oderl-acrobot": {"continuous": 0.82613117, "discrete": 3.5674138},
"oderl-cancer": {"continuous": 6.760299902695876, "discrete": 14.775529274573692},
},
}
def load_observing_var_threshold_ranges():
return {
0.1: {
"oderl-cartpole": {"continuous": {"lower": 0.0005, "upper": 0.45}, "discrete": 0.5}, # Use these ones!
"oderl-pendulum": {"continuous": {"lower": 0.0005, "upper": 0.45}, "discrete": 0.061973028},
"oderl-acrobot": {"continuous": {"lower": 0.02, "upper": 5.0}, "discrete": 0.413525},
"oderl-cancer": {"continuous": {"lower": 2.0, "upper": 19.0}, "discrete": 2.5688453},
},
0.4: {"oderl-cancer": {"continuous": {"lower": 2.0, "upper": 19.0}, "discrete": 14.775529274573692}},
}