-
Notifications
You must be signed in to change notification settings - Fork 6
Expand file tree
/
Copy pathdata_generation.py
More file actions
619 lines (534 loc) · 23.7 KB
/
Copy pathdata_generation.py
File metadata and controls
619 lines (534 loc) · 23.7 KB
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
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
# @file data_generation.py
# This file contains the code to generate a dataset for learning
#
# DataDir = "${CoverageControl_ws}/datasets/lpac" # Absolute location
# EnvironmentConfig = "${CoverageControl_ws}/datasets/lpac/coverage_control_params.toml" # Absolute location
#
# NumDataset = 1000
#
# # Number of steps to take before data is stores
# # This helps in creating a more diverse dataset
# EveryNumSteps = 5
#
# # The robots stop moving once the algorithm has converged
# # Having some of these converged steps can help in stabilizing robot actions
# ConvergedDataRatio = 0.02
#
# # Resizing of maps and Sparsification of tensors are triggered every TriggerPostProcessing dataset
# # This should be set based on RAM resources available on the system
# TriggerPostProcessing = 100
#
# CNNMapSize = 32
# EveryNumSteps = 5
# SaveObjective = false
# TimeStep = 5
# SaveRobotWorldLocalMaps = false
# SaveAsSparseQ = true
# NormalizeQ = true
#
# [DataSetSplit]
# TrainRatio = 0.7
# ValRatio = 0.2
# TestRatio = 0.1
#
# @file data_generation.py
# @brief Class to generate CoverageControl dataset for LPAC architecture
import datetime
import math
import os
import pathlib
import sys
import argparse
from distutils.util import strtobool
from rich.progress import (
Progress,
BarColumn,
TextColumn,
TimeRemainingColumn,
TimeElapsedColumn,
TaskProgressColumn,
MofNCompleteColumn,
)
import coverage_control
import torch
from coverage_control import CoverageSystem
from coverage_control import IOUtils
from coverage_control import CoverageEnvUtils
from coverage_control.algorithms import ClairvoyantCVT
from coverage_control.algorithms import CentralizedCVT
from coverage_control.algorithms import NearOptimalCVT
# @ingroup python_api
class DatasetGenerator:
"""
Class to generate CoverageControl dataset for LPAC architecture.
"""
def __init__(self, args):
self.config_file = args.config_file
self.config = IOUtils.load_toml(self.config_file)
self.split_dataset = bool(strtobool(args.split))
self.algorithm = "ClairvoyantCVT"
if args.algorithm:
print("Using CentralizedCVT algorithm")
self.algorithm = args.algorithm
self.data_dir = pathlib.Path(
IOUtils.sanitize_path(self.config["DataDir"]))
if not self.data_dir.exists():
print(f"{self.data_dir} does not exist")
sys.exit()
if args.append_dir:
self.dataset_dir = self.data_dir / args.append_dir
self.dataset_dir_path = pathlib.Path(self.dataset_dir)
if not self.dataset_dir_path.exists():
os.makedirs(self.dataset_dir)
env_config_file = IOUtils.sanitize_path(
self.config["EnvironmentConfig"])
env_config_file = pathlib.Path(env_config_file)
if not env_config_file.exists():
print(f"{env_config_file} does not exist")
sys.exit()
self.env_params = coverage_control.Parameters(
env_config_file.as_posix())
# Initialize variables
self.dataset_count = 0
self.non_converged_dataset_count = 0
self.converged_dataset_count = 0
self.env_count = 0
self.trigger_count = 0
self.trigger_start_idx = 0
self.num_dataset = self.config["NumDataset"]
self.converged_data_ratio = self.config["ConvergedDataRatio"]
self.num_converged_dataset = math.ceil(
self.converged_data_ratio * self.num_dataset
)
self.num_non_converged_dataset = self.num_dataset - self.num_converged_dataset
self.num_robots = self.env_params.pNumRobots
self.comm_range = self.env_params.pCommunicationRange
self.resolution = self.env_params.pResolution
self.cnn_map_size = self.config["CNNMapSize"]
self.every_num_step = self.config["EveryNumSteps"]
self.save_objective = False
if "SaveObjective" in self.config:
self.save_objective = self.config["SaveObjective"]
if "TimeStep" in self.config:
self.env_params.pTimeStep = self.config["TimeStep"]
self.save_robot_world_local_maps = False
if "SaveRobotWorldLocalMaps" in self.config:
self.save_robot_world_local_maps = self.config["SaveRobotWorldLocalMaps"]
self.trigger_size = self.config["TriggerPostProcessing"]
if self.trigger_size == 0 or self.trigger_size > self.num_dataset:
self.trigger_size = self.num_dataset
if torch.cuda.is_available():
self.device = torch.device("cuda")
else:
self.device = torch.device("cpu")
# Initialize tensors
self.actions = torch.zeros((self.num_dataset, self.num_robots, 2))
self.goals = torch.zeros((self.num_dataset, self.num_robots, 2))
self.robot_positions = torch.zeros(
(self.num_dataset, self.num_robots, 2))
self.raw_local_maps = torch.zeros(
(
self.trigger_size,
self.num_robots,
self.env_params.pLocalMapSize,
self.env_params.pLocalMapSize,
)
)
self.raw_obstacle_maps = torch.zeros(
(
self.trigger_size,
self.num_robots,
self.env_params.pLocalMapSize,
self.env_params.pLocalMapSize,
)
)
self.local_maps = torch.zeros(
(self.num_dataset, self.num_robots,
self.cnn_map_size, self.cnn_map_size)
)
self.obstacle_maps = torch.zeros(
(self.num_dataset, self.num_robots,
self.cnn_map_size, self.cnn_map_size)
)
self.comm_maps = torch.zeros(
(self.num_dataset, self.num_robots, 2,
self.cnn_map_size, self.cnn_map_size)
)
self.coverage_features = torch.zeros(
(self.num_dataset, self.num_robots, 7))
self.edge_weights = torch.zeros(
(self.num_dataset, self.num_robots, self.num_robots)
)
if self.save_objective:
self.objectives = torch.zeros(self.num_dataset)
if self.save_robot_world_local_maps:
self.raw_robot_world_local_maps = torch.zeros(
self.trigger_size,
self.num_robots,
self.env_params.pLocalMapSize,
self.env_params.pLocalMapSize,
)
self.robot_world_local_maps = torch.zeros(
self.num_dataset,
self.num_robots,
self.cnn_map_size,
self.cnn_map_size
)
self.start_time = datetime.datetime.now()
# Write metrics
self.metrics_file = self.dataset_dir_path / "metrics.txt"
# self.metrics = open(self.metrics_file, 'w')
with open(self.metrics_file, "w", encoding="utf-8") as f:
# f.write("Time: " + str(datetime.datetime.now()) + "\n")
f.write(f"Time: {self.start_time}\n")
f.write("Dataset directory: " + str(self.dataset_dir) + "\n")
self.print_tensor_sizes(f)
f.flush()
self.print_tensor_sizes()
columns = [
BarColumn(bar_width=None),
TaskProgressColumn(),
TextColumn("[progress.description]{task.description}"),
MofNCompleteColumn(),
TextColumn("#Envs: {task.fields[num_envs]}"),
TimeRemainingColumn(),
TimeElapsedColumn(),
]
with Progress(*columns, expand=True) as self.progress:
self.task = self.progress.add_task(
"[bold blue]Generating dataset",
total=self.num_dataset,
num_envs="",
)
self.run_data_generation()
self.save_dataset()
end_time = datetime.datetime.now()
with open(self.metrics_file, "a", encoding="utf-8") as f:
f.write("Time: " + str(datetime.datetime.now()) + "\n")
f.write("Total time: " + str(end_time - self.start_time) + "\n")
def run_data_generation(self):
num_non_converged_env = 0
while self.dataset_count < self.num_dataset:
self.env = CoverageSystem(self.env_params)
self.force_no_noise = True
if self.algorithm == "CentralizedCVT":
self.alg = CentralizedCVT(
self.env_params, self.num_robots, self.env, self.force_no_noise)
elif self.algorithm == "NearOptimalCVT":
self.alg = NearOptimalCVT(
self.env_params, self.num_robots, self.env, self.force_no_noise)
else:
self.alg = ClairvoyantCVT(
self.env_params, self.num_robots, self.env, self.force_no_noise)
self.env_count += 1
self.progress.update(
self.task,
num_envs=f"{self.env_count}"
)
self.progress.refresh()
num_steps = 0
is_converged = False
while (
num_steps < self.env_params.pEpisodeSteps
and not is_converged
and self.dataset_count < self.num_dataset
):
if num_steps % self.every_num_step == 0 and self.non_converged_dataset_count < self.num_non_converged_dataset:
is_converged = self.step_with_save()
self.progress.advance(self.task, advance=1)
self.non_converged_dataset_count += 1
else:
is_converged = self.step_without_save()
num_steps += 1
if num_steps == self.env_params.pEpisodeSteps:
num_non_converged_env += 1
if self.converged_dataset_count < self.num_converged_dataset and self.dataset_count < self.num_dataset:
self.step_with_save()
self.progress.advance(self.task, advance=1)
self.converged_dataset_count += 1
def step_with_save(self):
self.alg.ComputeActions()
converged = self.alg.IsConverged()
actions = self.alg.GetActions()
goals = self.alg.GetGoals()
robot_pos = self.env.GetRobotPositions()
for i, pos in enumerate(robot_pos):
goals[i] -= pos
count = self.dataset_count
self.actions[count] = CoverageEnvUtils.to_tensor(actions)
self.goals[count] = CoverageEnvUtils.to_tensor(goals)
self.robot_positions[count] = CoverageEnvUtils.get_robot_positions(
self.env)
self.coverage_features[count] = CoverageEnvUtils.get_voronoi_features(
self.env)
self.raw_local_maps[self.trigger_count] = CoverageEnvUtils.get_raw_local_maps(
self.env, self.env_params
)
self.raw_obstacle_maps[self.trigger_count] = (
CoverageEnvUtils.get_raw_obstacle_maps(self.env, self.env_params)
)
if self.save_robot_world_local_maps:
self.raw_robot_world_local_maps[self.trigger_count] = (
CoverageEnvUtils.get_raw_robot_world_local_maps(self.env, self.env_params)
)
self.comm_maps[count] = CoverageEnvUtils.get_communication_maps(
self.env, self.env_params, self.cnn_map_size
)
self.edge_weights[count] = CoverageEnvUtils.get_weights(
self.env, self.env_params
)
if self.save_objective:
self.objectives[count] = self.env.GetObjectiveValue()
self.dataset_count += 1
# if self.dataset_count % 100 == 0:
# print(f"Dataset: {self.dataset_count}/{self.num_dataset}")
# print(f"Elapsed time: {datetime.datetime.now() - self.start_time}")
self.trigger_count += 1
if self.trigger_count == self.trigger_size:
self.trigger_post_processing()
self.trigger_count = 0
error_flag = self.env.StepActions(actions)
return converged or error_flag
def trigger_post_processing(self):
if self.trigger_start_idx > self.num_dataset - 1:
return
trigger_end_idx = min(
self.num_dataset, self.trigger_start_idx + self.trigger_size
)
raw_local_maps = self.raw_local_maps[
0: trigger_end_idx - self.trigger_start_idx
]
raw_local_maps = raw_local_maps.to(self.device)
resized_local_maps = CoverageEnvUtils.resize_maps(
raw_local_maps, self.cnn_map_size
)
self.local_maps[self.trigger_start_idx: trigger_end_idx] = (
resized_local_maps.view(
-1, self.num_robots, self.cnn_map_size, self.cnn_map_size
)
.cpu()
.clone()
)
if self.save_robot_world_local_maps:
raw_robot_world_local_maps = self.raw_robot_world_local_maps[
0: trigger_end_idx - self.trigger_start_idx
]
raw_robot_world_local_maps = raw_robot_world_local_maps.to(self.device)
resized_robot_world_local_maps = CoverageEnvUtils.resize_maps(
raw_robot_world_local_maps, self.cnn_map_size
)
self.robot_world_local_maps[self.trigger_start_idx: trigger_end_idx] = (
resized_robot_world_local_maps.view(
-1, self.num_robots, self.cnn_map_size, self.cnn_map_size
)
.cpu()
.clone()
)
raw_obstacle_maps = self.raw_obstacle_maps[
0: trigger_end_idx - self.trigger_start_idx
]
raw_obstacle_maps = raw_obstacle_maps.to(self.device)
resized_obstacle_maps = CoverageEnvUtils.resize_maps(
raw_obstacle_maps, self.cnn_map_size
)
self.obstacle_maps[self.trigger_start_idx: trigger_end_idx] = (
resized_obstacle_maps.view(
-1, self.num_robots, self.cnn_map_size, self.cnn_map_size
)
.cpu()
.clone()
)
self.trigger_start_idx = trigger_end_idx
def normalize_tensor(self, tensor, epsilon=1e-6, zero_mean=False, is_symmetric=False):
dimensions = torch.tensor(range(tensor.dim()))[:-1]
if zero_mean is True:
tensor_mean = torch.zeros_like(dimensions)
else:
tensor_mean = tensor.mean(dim=list(dimensions))
tensor_std = tensor.std(dim=list(dimensions))
# Set tensor_std to be average of stds and keepdim=True to broadcast
if is_symmetric is True:
tensor_std = torch.ones_like(tensor_std) * tensor_std.mean()
if torch.isnan(tensor_std).any():
print("NaN in tensor std")
if torch.isnan(tensor_mean).any():
print("NaN in tensor mean")
# Check for division by zero and print warnin
if torch.any(tensor_std < epsilon):
print("Tensor: ", tensor_std)
print("normalize_tensor Warning: Division by zero in normalization")
print("Adding epsilon to std with zero values")
tensor_std = torch.where(tensor_std < epsilon, epsilon, tensor_std)
tensor = (tensor - tensor_mean) / tensor_std
return tensor, tensor_mean, tensor_std
def normalize_communication_maps(self):
min_val = self.comm_maps.min()
max_val = self.comm_maps.max()
range_val = max_val - min_val
self.comm_maps = (self.comm_maps - min_val) / range_val
print("Communication map min: " + str(min_val))
print("Communication map max: " + str(max_val))
return min_val, range_val
def save_tensor(self, tensor, name, as_sparse=False):
if self.split_dataset:
self.save_tensor_split(tensor, name, as_sparse)
else:
self.save_tensor_nosplit(tensor, name, as_sparse)
def save_tensor_nosplit(self, tensor, name, as_sparse=False):
tensor = tensor.cpu()
if as_sparse:
tensor = tensor.to_sparse()
dataset_dir_path = pathlib.Path(self.dataset_dir)
torch.save(tensor, dataset_dir_path / name)
def save_tensor_split(self, tensor, name, as_sparse=False):
tensor = tensor.cpu()
train_tensor = tensor[0: self.train_size].clone()
validation_tensor = tensor[
self.train_size: self.train_size + self.validation_size
].clone()
test_tensor = tensor[self.train_size + self.validation_size:].clone()
if as_sparse:
train_tensor = train_tensor.to_sparse()
validation_tensor = validation_tensor.to_sparse()
test_tensor = test_tensor.to_sparse()
dataset_dir_path = pathlib.Path(self.dataset_dir)
torch.save(train_tensor, dataset_dir_path / "train/" / name)
torch.save(validation_tensor, dataset_dir_path / "val/" / name)
torch.save(test_tensor, dataset_dir_path / "test/" / name)
def save_dataset(self):
as_sparse = self.config["SaveAsSparseQ"]
if not os.path.exists(self.dataset_dir):
os.makedirs(self.dataset_dir)
if self.split_dataset:
self.train_size = int(
self.num_dataset * self.config["DataSetSplit"]["TrainRatio"]
)
self.validation_size = int(
self.num_dataset * self.config["DataSetSplit"]["ValRatio"]
)
self.test_size = self.num_dataset - self.train_size - self.validation_size
# Make sure the folder exists
if not os.path.exists(self.dataset_dir / "train"):
os.makedirs(self.dataset_dir / "train")
if not os.path.exists(self.dataset_dir / "val"):
os.makedirs(self.dataset_dir / "val")
if not os.path.exists(self.dataset_dir / "test"):
os.makedirs(self.dataset_dir / "test")
self.save_tensor(self.robot_positions, "robot_positions.pt")
self.save_tensor(self.local_maps, "local_maps.pt", as_sparse)
if self.save_robot_world_local_maps:
self.save_tensor(self.robot_world_local_maps, "robot_world_local_maps.pt", as_sparse)
self.save_tensor(self.obstacle_maps, "obstacle_maps.pt", as_sparse)
self.save_tensor(self.edge_weights, "edge_weights.pt", as_sparse)
# min_val, range_val = self.normalize_communication_maps()
self.save_tensor(self.comm_maps, "comm_maps.pt", as_sparse)
# torch.save(min_val, self.dataset_dir / 'comm_maps_min.pt')
# torch.save(range_val, self.dataset_dir / 'comm_maps_range.pt')
self.save_tensor(self.actions, "actions.pt")
self.save_tensor(self.goals, "goals.pt")
self.save_tensor(self.coverage_features, "coverage_features.pt")
if self.save_objective:
self.save_tensor(self.objectives, "objectives.pt")
if self.config["NormalizeQ"]:
normalized_actions, actions_mean, actions_std = self.normalize_tensor(
self.actions, is_symmetric=True, zero_mean=True
)
normalized_goals, goals_mean, goals_std = self.normalize_tensor(
self.goals, is_symmetric=True, zero_mean=True
)
self.save_tensor(normalized_actions, "normalized_actions.pt")
torch.save(actions_mean, self.dataset_dir_path / "actions_mean.pt")
torch.save(actions_std, self.dataset_dir_path / "actions_std.pt")
self.save_tensor(normalized_goals, "normalized_goals.pt")
torch.save(goals_mean, self.dataset_dir_path / "goals_mean.pt")
torch.save(goals_std, self.dataset_dir_path / "goals_std.pt")
if self.save_objective:
normalize_objectives, objectives_mean, objectives_std = self.normalize_tensor(
self.objectives
)
self.save_tensor(normalize_objectives, "normalized_objectives.pt")
torch.save(
objectives_mean, self.dataset_dir_path / "objectives_mean.pt"
)
torch.save(
objectives_std, self.dataset_dir_path / "objectives_std.pt"
)
# coverage_features, coverage_features_mean, coverage_features_std = (
# self.normalize_tensor(self.coverage_features)
# )
# self.save_tensor(coverage_features, "normalized_coverage_features.pt")
# torch.save(
# coverage_features_mean,
# self.dataset_dir_path / "coverage_features_mean.pt",
# )
# torch.save(
# coverage_features_std,
# self.dataset_dir_path / "coverage_features_std.pt",
# )
def step_without_save(self):
self.alg.ComputeActions()
converged = self.alg.IsConverged()
if self.env.StepActions(self.alg.GetActions()):
return True
return converged
def get_tensor_byte_size_mb(self, tensor):
return (tensor.element_size() * tensor.nelement()) / (1024 * 1024)
def print_tensor_sizes(self, file=sys.stdout):
# Set to two decimal places
print("Tensor sizes:", file=file)
print("Actions:", self.get_tensor_byte_size_mb(self.actions), file=file)
print("Goals:", self.get_tensor_byte_size_mb(self.goals), file=file)
print(
"Robot positions:",
self.get_tensor_byte_size_mb(self.robot_positions),
file=file,
)
print(
"Raw local maps:",
self.get_tensor_byte_size_mb(self.raw_local_maps),
file=file,
)
print(
"Raw obstacle maps:",
self.get_tensor_byte_size_mb(self.raw_obstacle_maps),
file=file,
)
print("Local maps:", self.get_tensor_byte_size_mb(
self.local_maps), file=file)
if self.save_robot_world_local_maps:
print("Robot world local maps: ", self.get_tensor_byte_size_mb(
self.robot_world_local_maps), file=file)
print(
"Obstacle maps:",
self.get_tensor_byte_size_mb(self.obstacle_maps),
file=file,
)
print("Comm maps:", self.get_tensor_byte_size_mb(
self.comm_maps), file=file)
print(
"Coverage features:",
self.get_tensor_byte_size_mb(self.coverage_features),
file=file,
)
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument("config_file", type=str, help="Path to config file")
parser.add_argument("--append-dir",
type=str,
default="data",
help="Append directory to dataset path",
required=False
)
parser.add_argument("--split",
type=str,
default="False",
help="Split dataset into train, validation and test sets",
required=False
)
parser.add_argument("--algorithm",
type=str,
help="Algorithm for generating dataset",
required=False,
)
args = parser.parse_args()
DatasetGenerator(args)