Spaces:
Sleeping
Sleeping
File size: 15,505 Bytes
eaf2e33 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 |
from collections import OrderedDict
import numpy as np
import torch
from src.rlkit.data_management.replay_buffer import ReplayBuffer, EnsembleReplayBuffer
class SimpleReplayBuffer(ReplayBuffer):
def __init__(
self,
max_replay_buffer_size,
observation_dim,
action_dim,
env_info_sizes,
):
self._observation_dim = observation_dim
self._action_dim = action_dim
self._max_replay_buffer_size = max_replay_buffer_size
self._observations = np.zeros((max_replay_buffer_size, observation_dim))
# It's a bit memory inefficient to save the observations twice,
# but it makes the code *much* easier since you no longer have to
# worry about termination conditions.
self._next_obs = np.zeros((max_replay_buffer_size, observation_dim))
self._actions = np.zeros((max_replay_buffer_size, action_dim))
# Make everything a 2D np array to make it easier for other code to
# reason about the shape of the data
self._rewards = np.zeros((max_replay_buffer_size, 1))
# self._terminals[i] = a terminal was received at time i
self._terminals = np.zeros((max_replay_buffer_size, 1), dtype='uint8')
# Define self._env_infos[key][i] to be the return value of env_info[key]
# at time i
self._env_infos = {}
for key, size in env_info_sizes.items():
self._env_infos[key] = np.zeros((max_replay_buffer_size, size))
self._env_info_keys = env_info_sizes.keys()
self._top = 0
self._size = 0
def add_sample(self, observation, action, reward, next_observation,
terminal, env_info, **kwargs):
self._observations[self._top] = observation
self._actions[self._top] = action
self._rewards[self._top] = reward
self._terminals[self._top] = terminal
self._next_obs[self._top] = next_observation
for key in self._env_info_keys:
self._env_infos[key][self._top] = env_info[key]
self._advance()
def terminate_episode(self):
pass
def _advance(self):
self._top = (self._top + 1) % self._max_replay_buffer_size
if self._size < self._max_replay_buffer_size:
self._size += 1
def random_batch(self, batch_size):
indices = np.random.randint(0, self._size, batch_size)
batch = dict(
observations=self._observations[indices],
actions=self._actions[indices],
rewards=self._rewards[indices],
terminals=self._terminals[indices],
next_observations=self._next_obs[indices],
)
for key in self._env_info_keys:
assert key not in batch.keys()
batch[key] = self._env_infos[key][indices]
return batch
def rebuild_env_info_dict(self, idx):
return {
key: self._env_infos[key][idx]
for key in self._env_info_keys
}
def batch_env_info_dict(self, indices):
return {
key: self._env_infos[key][indices]
for key in self._env_info_keys
}
def num_steps_can_sample(self):
return self._size
def get_diagnostics(self):
return OrderedDict([
('size', self._size)
])
class EnsembleSimpleReplayBuffer(EnsembleReplayBuffer):
def __init__(
self,
max_replay_buffer_size,
observation_dim,
action_dim,
env_info_sizes,
num_ensemble,
log_dir,
):
self._observation_dim = observation_dim
self._action_dim = action_dim
self._max_replay_buffer_size = max_replay_buffer_size
self._observations = np.zeros((max_replay_buffer_size, observation_dim))
# It's a bit memory inefficient to save the observations twice,
# but it makes the code *much* easier since you no longer have to
# worry about termination conditions.
self._next_obs = np.zeros((max_replay_buffer_size, observation_dim))
self._actions = np.zeros((max_replay_buffer_size, action_dim))
# Make everything a 2D np array to make it easier for other code to
# reason about the shape of the data
self._rewards = np.zeros((max_replay_buffer_size, 1))
# self._terminals[i] = a terminal was received at time i
self._terminals = np.zeros((max_replay_buffer_size, 1), dtype='uint8')
# Define self._env_infos[key][i] to be the return value of env_info[key]
# at time i
self._env_infos = {}
for key, size in env_info_sizes.items():
self._env_infos[key] = np.zeros((max_replay_buffer_size, size))
self._env_info_keys = env_info_sizes.keys()
# define mask
self._mask = np.zeros((max_replay_buffer_size, num_ensemble))
self._top = 0
self._size = 0
self.buffer_dir = log_dir + '/buffer/'
def add_sample(self, observation, action, reward, next_observation,
terminal, mask, env_info, **kwargs):
self._observations[self._top] = observation
self._actions[self._top] = action
self._rewards[self._top] = reward
self._terminals[self._top] = terminal
self._next_obs[self._top] = next_observation
self._mask[self._top] = mask
for key in self._env_info_keys:
self._env_infos[key][self._top] = env_info[key]
self._advance()
def terminate_episode(self):
pass
def _advance(self):
self._top = (self._top + 1) % self._max_replay_buffer_size
if self._size < self._max_replay_buffer_size:
self._size += 1
def random_batch(self, batch_size):
indices = np.random.randint(0, self._size, batch_size)
batch = dict(
observations=self._observations[indices],
actions=self._actions[indices],
rewards=self._rewards[indices],
terminals=self._terminals[indices],
next_observations=self._next_obs[indices],
masks=self._mask[indices],
)
for key in self._env_info_keys:
assert key not in batch.keys()
batch[key] = self._env_infos[key][indices]
return batch
def rebuild_env_info_dict(self, idx):
return {
key: self._env_infos[key][idx]
for key in self._env_info_keys
}
def batch_env_info_dict(self, indices):
return {
key: self._env_infos[key][indices]
for key in self._env_info_keys
}
def num_steps_can_sample(self):
return self._size
def get_diagnostics(self):
return OrderedDict([
('size', self._size)
])
def save_buffer(self, epoch):
path = self.buffer_dir + '/replay_%d.pt' % (epoch)
payload = [
self._observations[:self._size],
self._actions[:self._size],
self._rewards[:self._size],
self._terminals[:self._size],
self._next_obs[:self._size],
self._mask[:self._size],
self._size,
]
torch.save(payload, path)
def load_buffer(self, epoch):
path = self.buffer_dir + '/replay_%d.pt' % (epoch)
payload = torch.load(path)
self._observations = payload[0]
self._actions = payload[1]
self._rewards = payload[2]
self._terminals = payload[3]
self._next_obs = payload[4]
self._mask = payload[5]
self._size = payload[6]
class RandomReplayBuffer(ReplayBuffer):
def __init__(
self,
max_replay_buffer_size,
observation_dim,
action_dim,
env_info_sizes,
single_flag,
equal_flag,
lower,
upper,
):
self._observation_dim = observation_dim
self._action_dim = action_dim
self._max_replay_buffer_size = max_replay_buffer_size
self._observations = np.zeros((max_replay_buffer_size, observation_dim))
# It's a bit memory inefficient to save the observations twice,
# but it makes the code *much* easier since you no longer have to
# worry about termination conditions.
self._next_obs = np.zeros((max_replay_buffer_size, observation_dim))
self._actions = np.zeros((max_replay_buffer_size, action_dim))
# Make everything a 2D np array to make it easier for other code to
# reason about the shape of the data
self._rewards = np.zeros((max_replay_buffer_size, 1))
# self._terminals[i] = a terminal was received at time i
self._terminals = np.zeros((max_replay_buffer_size, 1), dtype='uint8')
# Define self._env_infos[key][i] to be the return value of env_info[key]
# at time i
self._env_infos = {}
for key, size in env_info_sizes.items():
self._env_infos[key] = np.zeros((max_replay_buffer_size, size))
self._env_info_keys = env_info_sizes.keys()
self._top = 0
self._size = 0
# randomization
self.single_flag = single_flag
self.equal_flag = equal_flag
self.lower = lower
self.upper = upper
def add_sample(self, observation, action, reward, next_observation,
terminal, env_info, **kwargs):
self._observations[self._top] = observation
self._actions[self._top] = action
self._rewards[self._top] = reward
self._terminals[self._top] = terminal
self._next_obs[self._top] = next_observation
for key in self._env_info_keys:
self._env_infos[key][self._top] = env_info[key]
self._advance()
def terminate_episode(self):
pass
def _advance(self):
self._top = (self._top + 1) % self._max_replay_buffer_size
if self._size < self._max_replay_buffer_size:
self._size += 1
def random_batch(self, batch_size):
indices = np.random.randint(0, self._size, batch_size)
obs = self._observations[indices]
next_obs = self._next_obs[indices]
if self.single_flag == 0:
random_number_1 = np.random.uniform(self.lower, self.upper, obs.shape[0]).reshape(-1,1)
random_number_2 = np.random.uniform(self.lower, self.upper, obs.shape[0]).reshape(-1,1)
else:
random_number_1 = np.random.uniform(self.lower, self.upper, obs.shape[0]*obs.shape[1]).reshape(obs.shape[0],-1)
random_number_2 = np.random.uniform(self.lower, self.upper, obs.shape[0]*obs.shape[1]).reshape(obs.shape[0],-1)
if self.equal_flag == 0:
obs = obs * random_number_1
next_obs = next_obs * random_number_1
else:
obs = obs * random_number_1
next_obs = next_obs * random_number_2
batch = dict(
observations=obs,
actions=self._actions[indices],
rewards=self._rewards[indices],
terminals=self._terminals[indices],
next_observations=next_obs,
)
for key in self._env_info_keys:
assert key not in batch.keys()
batch[key] = self._env_infos[key][indices]
return batch
def rebuild_env_info_dict(self, idx):
return {
key: self._env_infos[key][idx]
for key in self._env_info_keys
}
def batch_env_info_dict(self, indices):
return {
key: self._env_infos[key][indices]
for key in self._env_info_keys
}
def num_steps_can_sample(self):
return self._size
def get_diagnostics(self):
return OrderedDict([
('size', self._size)
])
class GaussianReplayBuffer(ReplayBuffer):
def __init__(
self,
max_replay_buffer_size,
observation_dim,
action_dim,
env_info_sizes,
prob,
std,
):
self._observation_dim = observation_dim
self._action_dim = action_dim
self._max_replay_buffer_size = max_replay_buffer_size
self._observations = np.zeros((max_replay_buffer_size, observation_dim))
# It's a bit memory inefficient to save the observations twice,
# but it makes the code *much* easier since you no longer have to
# worry about termination conditions.
self._next_obs = np.zeros((max_replay_buffer_size, observation_dim))
self._actions = np.zeros((max_replay_buffer_size, action_dim))
# Make everything a 2D np array to make it easier for other code to
# reason about the shape of the data
self._rewards = np.zeros((max_replay_buffer_size, 1))
# self._terminals[i] = a terminal was received at time i
self._terminals = np.zeros((max_replay_buffer_size, 1), dtype='uint8')
# Define self._env_infos[key][i] to be the return value of env_info[key]
# at time i
self._env_infos = {}
for key, size in env_info_sizes.items():
self._env_infos[key] = np.zeros((max_replay_buffer_size, size))
self._env_info_keys = env_info_sizes.keys()
self._top = 0
self._size = 0
# randomization
self.prob = prob
self.std = std
def add_sample(self, observation, action, reward, next_observation,
terminal, env_info, **kwargs):
self._observations[self._top] = observation
self._actions[self._top] = action
self._rewards[self._top] = reward
self._terminals[self._top] = terminal
self._next_obs[self._top] = next_observation
for key in self._env_info_keys:
self._env_infos[key][self._top] = env_info[key]
self._advance()
def terminate_episode(self):
pass
def _advance(self):
self._top = (self._top + 1) % self._max_replay_buffer_size
if self._size < self._max_replay_buffer_size:
self._size += 1
def random_batch(self, batch_size):
indices = np.random.randint(0, self._size, batch_size)
obs = self._observations[indices]
next_obs = self._next_obs[indices]
num_batch, dim_input = obs.shape[0], obs.shape[1]
noise = np.random.normal(0, self.std, num_batch*dim_input).reshape(num_batch, -1)
mask = np.random.uniform(0, 1, num_batch).reshape(num_batch, -1) < self.prob
noise = noise * mask
obs = obs + noise
next_obs = next_obs + noise
batch = dict(
observations=obs,
actions=self._actions[indices],
rewards=self._rewards[indices],
terminals=self._terminals[indices],
next_observations=next_obs,
)
for key in self._env_info_keys:
assert key not in batch.keys()
batch[key] = self._env_infos[key][indices]
return batch
def rebuild_env_info_dict(self, idx):
return {
key: self._env_infos[key][idx]
for key in self._env_info_keys
}
def batch_env_info_dict(self, indices):
return {
key: self._env_infos[key][indices]
for key in self._env_info_keys
}
def num_steps_can_sample(self):
return self._size
def get_diagnostics(self):
return OrderedDict([
('size', self._size)
])
|