-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrain_data_sharing.py
189 lines (155 loc) · 8.06 KB
/
train_data_sharing.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
import os
from typing import Tuple
import gym
import numpy as np
import tqdm
from absl import app, flags
from ml_collections import config_flags
from tensorboardX import SummaryWriter
import wrappers
from dataset_utils import D4RLDataset, split_into_trajectories
from evaluation import evaluate
from learner import Learner
import time
import preprocessing
from preprocessing import learned_reward
from preprocessing.learned_reward import merge_dataset
FLAGS = flags.FLAGS
flags.DEFINE_string('env_name', 'halfcheetah-expert-v2', 'Environment name.')
flags.DEFINE_string('source_name', 'halfcheetah-random-v2', 'Source Environment name.')
flags.DEFINE_string('comment', 'data_sharing', 'Comment for the run')
flags.DEFINE_string('save_dir', './runs/', 'Tensorboard logging dir.')
flags.DEFINE_string('data_share', 'none', 'share type.')
flags.DEFINE_float('target_split', 1, 'amount of target data to use')
flags.DEFINE_float('source_split', 1, 'amount of target data to use')
flags.DEFINE_float('weight', 10, 'std weight')
flags.DEFINE_float('ensemble', 10, 'ensemble')
# flags.DEFINE_integer('seed', 42, 'Random seed.')
flags.DEFINE_integer('seed', int(time.time()), 'Random seed.')
flags.DEFINE_integer('eval_episodes', 10,
'Number of episodes used for evaluation.')
flags.DEFINE_integer('log_interval', 5000, 'Logging interval.')
flags.DEFINE_integer('eval_interval', 5000, 'Eval interval.')
flags.DEFINE_integer('batch_size', 256, 'Mini batch size.')
flags.DEFINE_integer('max_steps', int(1e6), 'Number of training steps.')
flags.DEFINE_boolean('tqdm', True, 'Use tqdm progress bar.')
config_flags.DEFINE_config_file(
'config',
'default.py',
'File path to the training hyperparameter configuration.',
lock_config=False)
def normalize(dataset, split=1.):
trajs = split_into_trajectories(dataset.observations, dataset.actions,
dataset.rewards, dataset.masks,
dataset.dones_float,
dataset.next_observations)
def compute_returns(traj):
episode_return = 0
for _, _, rew, _, _, _ in traj:
episode_return += rew
return episode_return
trajs.sort(key=compute_returns)
dataset.rewards /= compute_returns(trajs[-1]) - compute_returns(trajs[0])
dataset.rewards *= 1000.0
if split < 1:
assert split > 0
sample_index = np.random.choice(np.arange(dataset.size), int(dataset.size * split), replace=False)
dataset.rewards = dataset.rewards[sample_index, ...]
dataset.observations = dataset.observations[sample_index, ...]
dataset.next_observations = dataset.next_observations[sample_index, ...]
dataset.actions = dataset.actions[sample_index, ...]
dataset.dones_float = dataset.dones_float[sample_index, ...]
dataset.masks = dataset.masks[sample_index, ...]
dataset.size = int(dataset.size * split)
def make_env_and_dataset(env_name: str, target_split: int, source_env_name: str, source_split: int,
seed: int, writer: any, variant) -> Tuple[gym.Env, D4RLDataset]:
env = gym.make(env_name)
env = wrappers.EpisodeMonitor(env)
env = wrappers.SinglePrecision(env)
env.seed(seed)
env.action_space.seed(seed)
env.observation_space.seed(seed)
source_env = gym.make(source_env_name)
source_dataset = D4RLDataset(source_env)
dataset = D4RLDataset(env)
if 'antmaze' in FLAGS.env_name:
dataset.rewards -= 1.0
source_dataset.rewards -= 1.0
# See https://github.com/aviralkumar2907/CQL/blob/master/d4rl/examples/cql_antmaze_new.py#L22
# but I found no difference between (x - 0.5) * 4 and x - 1.0
elif ('halfcheetah' in FLAGS.env_name or 'walker2d' in FLAGS.env_name
or 'hopper' in FLAGS.env_name):
normalize(dataset, target_split)
normalize(source_dataset, source_split)
print('target dataset: ', dataset.rewards.shape, 'other data: ', source_dataset.rewards.shape, '\n', '='*30)
merge_dataset(dataset, source_dataset, FLAGS.data_share, writer, variant)
return env, dataset
def main(_):
hidden_dim = preprocessing.training_reward.MIDDLE_SHAPE
variant = {'weight_std': FLAGS.weight, 'hidden_dim': hidden_dim}
# >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
# hidden_add:
# summary_writer = SummaryWriter(os.path.join(FLAGS.save_dir,
# f"{str(variant['weight_std'])}_hidden_add_{str(variant['hidden_dim'])}_{FLAGS.data_share}_{FLAGS.env_name}_{FLAGS.source_name}_{str(FLAGS.seed)}"),
# write_to_disk=True)
# >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
# learn_ensemble:
# summary_writer = SummaryWriter(os.path.join(FLAGS.save_dir,
# f"{str(variant['weight_std'])}_{FLAGS.data_share}_{FLAGS.env_name}_{FLAGS.source_name}_{str(FLAGS.seed)}"),
# write_to_disk=True)
# >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
# early_stopping:
# train_epoch = learned_reward.NUM_EPOCHS
# summary_writer = SummaryWriter(os.path.join(FLAGS.save_dir,
# f"early_stop_{str(variant['weight_std'])}_train_epoch_{str(train_epoch)}_{FLAGS.data_share}_{FLAGS.env_name}_{FLAGS.source_name}_{str(FLAGS.seed)}"),
# write_to_disk=True)
# >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
# L_min_std:
# summary_writer = SummaryWriter(os.path.join(FLAGS.save_dir,
# f"L_min_ensemble_{str(variant['weight_std'])}_{FLAGS.data_share}_{FLAGS.env_name}_{FLAGS.source_name}_{str(FLAGS.seed)}"),
# write_to_disk=True)
# >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
# L_min source splits
# learned_reward.ENSEMBLE = int(FLAGS.ensemble)
# summary_writer = SummaryWriter(os.path.join(FLAGS.save_dir,
# f"L_min_ensemble_{str(FLAGS.ensemble)}_{FLAGS.data_share}_{FLAGS.env_name}_{FLAGS.source_name}_source_split_{str(FLAGS.source_split)}_{str(FLAGS.seed)}"),
# write_to_disk=True)
# >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
# L_min delta
summary_writer = SummaryWriter(os.path.join(FLAGS.save_dir,
f"L_min_delta_{FLAGS.data_share}_{FLAGS.env_name}_{FLAGS.source_name}_source_split_{str(FLAGS.source_split)}_{str(FLAGS.seed)}"),
write_to_disk=True)
# >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
# summary_writer = SummaryWriter(os.path.join(FLAGS.save_dir,
# f"{str(variant['weight_std'])}_{FLAGS.data_share}_{FLAGS.env_name}_{FLAGS.source_name}_{str(FLAGS.seed)}"),
# write_to_disk=True)
os.makedirs(FLAGS.save_dir, exist_ok=True)
env, dataset = make_env_and_dataset(FLAGS.env_name, FLAGS.target_split, FLAGS.source_name, FLAGS.source_split,
FLAGS.seed, summary_writer, variant)
kwargs = dict(FLAGS.config)
agent = Learner(FLAGS.seed,
env.observation_space.sample()[np.newaxis],
env.action_space.sample()[np.newaxis],
max_steps=FLAGS.max_steps,
**kwargs)
eval_returns = []
for i in tqdm.tqdm(range(1, FLAGS.max_steps + 1),
smoothing=0.1,
disable=not FLAGS.tqdm):
batch = dataset.sample(FLAGS.batch_size)
update_info = agent.update(batch)
if i % FLAGS.log_interval == 0:
for k, v in update_info.items():
if v.ndim == 0:
summary_writer.add_scalar(f'training/{k}', v, i)
else:
summary_writer.add_histogram(f'training/{k}', v, i)
summary_writer.flush()
if i % FLAGS.eval_interval == 0:
eval_stats = evaluate(agent, env, FLAGS.eval_episodes)
for k, v in eval_stats.items():
summary_writer.add_scalar(f'evaluation/average_{k}s', v, i)
summary_writer.flush()
eval_returns.append((i, eval_stats['return']))
if __name__ == '__main__':
app.run(main)