-
Notifications
You must be signed in to change notification settings - Fork 112
/
pruning_imagenet.py
683 lines (585 loc) · 29.4 KB
/
pruning_imagenet.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
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
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
# https://github.com/pytorch/vision/blob/master/torchvision/models/__init__.py
import argparse
import os, sys
import shutil
import time
import torch
import torch.nn as nn
import torch.nn.parallel
import torch.backends.cudnn as cudnn
import torch.optim
import torch.utils.data
import torchvision.transforms as transforms
import torchvision.datasets as datasets
import torchvision.models
from utils import convert_secs2time, time_string, time_file_str, timing
# from models import print_log
import models
import random
import numpy as np
from scipy.spatial import distance
from collections import OrderedDict
model_names = sorted(name for name in models.__dict__
if name.islower() and not name.startswith("__")
and callable(models.__dict__[name]))
parser = argparse.ArgumentParser(description='PyTorch ImageNet Training')
parser.add_argument('data', metavar='DIR',
help='path to dataset')
parser.add_argument('--save_dir', type=str, default='./', help='Folder to save checkpoints and log.')
parser.add_argument('--arch', '-a', metavar='ARCH', default='resnet18',
choices=model_names,
help='model architecture: ' +
' | '.join(model_names) +
' (default: resnet18)')
parser.add_argument('-j', '--workers', default=4, type=int, metavar='N',
help='number of data loading workers (default: 4)')
parser.add_argument('--epochs', default=100, type=int, metavar='N', help='number of total epochs to run')
parser.add_argument('--start-epoch', default=0, type=int, metavar='N', help='manual epoch number (useful on restarts)')
parser.add_argument('-b', '--batch-size', default=256, type=int, metavar='N', help='mini-batch size (default: 256)')
parser.add_argument('--lr', '--learning-rate', default=0.1, type=float, metavar='LR', help='initial learning rate')
parser.add_argument('--momentum', default=0.9, type=float, metavar='M', help='momentum')
parser.add_argument('--weight-decay', '--wd', default=1e-4, type=float, metavar='W',
help='weight decay (default: 1e-4)')
parser.add_argument('--print-freq', '-p', default=200, type=int, metavar='N', help='print frequency (default: 100)')
parser.add_argument('--resume', default='', type=str, metavar='PATH', help='path to latest checkpoint (default: none)')
parser.add_argument('-e', '--evaluate', dest='evaluate', action='store_true', help='evaluate model on validation set')
parser.add_argument('--use_pretrain', dest='use_pretrain', action='store_true', help='use pre-trained model or not')
# compress rate
parser.add_argument('--rate_norm', type=float, default=0.9, help='the remaining ratio of pruning based on Norm')
parser.add_argument('--rate_dist', type=float, default=0.1, help='the reducing ratio of pruning based on Distance')
parser.add_argument('--layer_begin', type=int, default=3, help='compress layer of model')
parser.add_argument('--layer_end', type=int, default=3, help='compress layer of model')
parser.add_argument('--layer_inter', type=int, default=1, help='compress layer of model')
parser.add_argument('--epoch_prune', type=int, default=1, help='epoch interval of pruning')
parser.add_argument('--skip_downsample', type=int, default=1, help='compress layer of model')
parser.add_argument('--use_sparse', dest='use_sparse', action='store_true', help='use sparse model as initial or not')
parser.add_argument('--sparse',
default='/data/yahe/imagenet/resnet50-rate-0.7/checkpoint.resnet50.2018-01-07-9744.pth.tar',
type=str, metavar='PATH', help='path of sparse model')
parser.add_argument('--lr_adjust', type=int, default=30, help='number of epochs that change learning rate')
parser.add_argument('--VGG_pruned_style', choices=["CP_5x", "Thinet_conv"],
help='number of epochs that change learning rate')
args = parser.parse_args()
args.use_cuda = torch.cuda.is_available()
args.prefix = time_file_str()
def main():
best_prec1 = 0
if not os.path.isdir(args.save_dir):
os.makedirs(args.save_dir)
log = open(os.path.join(args.save_dir, '{}.{}.log'.format(args.arch, args.prefix)), 'w')
# version information
print_log("PyThon version : {}".format(sys.version.replace('\n', ' ')), log)
print_log("PyTorch version : {}".format(torch.__version__), log)
print_log("cuDNN version : {}".format(torch.backends.cudnn.version()), log)
print_log("Vision version : {}".format(torchvision.__version__), log)
# create model
print_log("=> creating model '{}'".format(args.arch), log)
model = models.__dict__[args.arch](pretrained=args.use_pretrain)
if args.use_sparse:
model = import_sparse(model)
print_log("=> Model : {}".format(model), log)
print_log("=> parameter : {}".format(args), log)
print_log("Norm Pruning Rate: {}".format(args.rate_norm), log)
print_log("Distance Pruning Rate: {}".format(args.rate_dist), log)
print_log("Layer Begin: {}".format(args.layer_begin), log)
print_log("Layer End: {}".format(args.layer_end), log)
print_log("Layer Inter: {}".format(args.layer_inter), log)
print_log("Epoch prune: {}".format(args.epoch_prune), log)
print_log("Skip downsample : {}".format(args.skip_downsample), log)
print_log("Workers : {}".format(args.workers), log)
print_log("Learning-Rate : {}".format(args.lr), log)
print_log("Use Pre-Trained : {}".format(args.use_pretrain), log)
print_log("lr adjust : {}".format(args.lr_adjust), log)
print_log("VGG pruned style : {}".format(args.VGG_pruned_style), log)
if args.arch.startswith('alexnet') or args.arch.startswith('vgg'):
model.features = torch.nn.DataParallel(model.features)
model.cuda()
else:
model = torch.nn.DataParallel(model).cuda()
# define loss function (criterion) and optimizer
criterion = nn.CrossEntropyLoss().cuda()
optimizer = torch.optim.SGD(model.parameters(), args.lr,
momentum=args.momentum,
weight_decay=args.weight_decay,
nesterov=True)
# optionally resume from a checkpoint
if args.resume:
if os.path.isfile(args.resume):
print_log("=> loading checkpoint '{}'".format(args.resume), log)
checkpoint = torch.load(args.resume)
args.start_epoch = checkpoint['epoch']
best_prec1 = checkpoint['best_prec1']
model.load_state_dict(checkpoint['state_dict'])
optimizer.load_state_dict(checkpoint['optimizer'])
print_log("=> loaded checkpoint '{}' (epoch {})".format(args.resume, checkpoint['epoch']), log)
else:
print_log("=> no checkpoint found at '{}'".format(args.resume), log)
cudnn.benchmark = True
# Data loading code
traindir = os.path.join(args.data, 'train')
valdir = os.path.join(args.data, 'val')
normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
train_dataset = datasets.ImageFolder(
traindir,
transforms.Compose([
transforms.RandomResizedCrop(224),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
normalize,
]))
train_loader = torch.utils.data.DataLoader(
train_dataset, batch_size=args.batch_size, shuffle=True,
num_workers=args.workers, pin_memory=True, sampler=None)
val_loader = torch.utils.data.DataLoader(
datasets.ImageFolder(valdir, transforms.Compose([
transforms.Resize(256),
transforms.CenterCrop(224),
transforms.ToTensor(),
normalize,
])),
batch_size=args.batch_size, shuffle=False,
num_workers=args.workers, pin_memory=True)
if args.evaluate:
validate(val_loader, model, criterion, log)
return
filename = os.path.join(args.save_dir, 'checkpoint.{:}.{:}.pth.tar'.format(args.arch, args.prefix))
bestname = os.path.join(args.save_dir, 'best.{:}.{:}.pth.tar'.format(args.arch, args.prefix))
m = Mask(model)
m.init_length()
print("-" * 10 + "one epoch begin" + "-" * 10)
print("remaining ratio of pruning : Norm is %f" % args.rate_norm)
print("reducing ratio of pruning : Distance is %f" % args.rate_dist)
print("total remaining ratio is %f" % (args.rate_norm - args.rate_dist))
m.model = model
m.init_mask(args.rate_norm, args.rate_dist)
# m.if_zero()
m.do_mask()
m.do_similar_mask()
model = m.model
m.if_zero()
if args.use_cuda:
model = model.cuda()
val_acc_2 = validate(val_loader, model, criterion, log)
print(">>>>> accu after is: {:}".format(val_acc_2))
start_time = time.time()
epoch_time = AverageMeter()
for epoch in range(args.start_epoch, args.epochs):
adjust_learning_rate(optimizer, epoch)
need_hour, need_mins, need_secs = convert_secs2time(epoch_time.val * (args.epochs - epoch))
need_time = '[Need: {:02d}:{:02d}:{:02d}]'.format(need_hour, need_mins, need_secs)
print_log(
' [{:s}] :: {:3d}/{:3d} ----- [{:s}] {:s}'.format(args.arch, epoch, args.epochs, time_string(), need_time),
log)
# train for one epoch
train(train_loader, model, criterion, optimizer, epoch, log, m)
# evaluate on validation set
val_acc_1 = validate(val_loader, model, criterion, log)
if epoch % args.epoch_prune == 0 or epoch == args.epochs - 1:
m.model = model
m.if_zero()
m.init_mask(args.rate_norm, args.rate_dist)
m.do_mask()
m.do_similar_mask()
m.if_zero()
model = m.model
if args.use_cuda:
model = model.cuda()
val_acc_2 = validate(val_loader, model, criterion, log)
# remember best prec@1 and save checkpoint
is_best = val_acc_2 > best_prec1
best_prec1 = max(val_acc_2, best_prec1)
save_checkpoint({
'epoch': epoch + 1,
'arch': args.arch,
'state_dict': model.state_dict(),
'best_prec1': best_prec1,
'optimizer': optimizer.state_dict(),
}, is_best, filename, bestname)
# measure elapsed time
epoch_time.update(time.time() - start_time)
start_time = time.time()
log.close()
def import_sparse(model):
checkpoint = torch.load(args.sparse)
new_state_dict = OrderedDict()
for k, v in checkpoint['state_dict'].items():
name = k[7:] # remove `module.`
new_state_dict[name] = v
model.load_state_dict(new_state_dict)
print("sparse_model_loaded")
return model
def train(train_loader, model, criterion, optimizer, epoch, log, m):
batch_time = AverageMeter()
data_time = AverageMeter()
losses = AverageMeter()
top1 = AverageMeter()
top5 = AverageMeter()
# switch to train mode
model.train()
end = time.time()
for i, (input, target) in enumerate(train_loader):
# measure data loading time
data_time.update(time.time() - end)
target = target.cuda(async=True)
input_var = torch.autograd.Variable(input)
target_var = torch.autograd.Variable(target)
# compute output
output = model(input_var)
loss = criterion(output, target_var)
# measure accuracy and record loss
prec1, prec5 = accuracy(output.data, target, topk=(1, 5))
losses.update(loss.data[0], input.size(0))
top1.update(prec1[0], input.size(0))
top5.update(prec5[0], input.size(0))
# compute gradient and do SGD step
optimizer.zero_grad()
loss.backward()
# Mask grad for iteration
m.do_grad_mask()
optimizer.step()
# measure elapsed time
batch_time.update(time.time() - end)
end = time.time()
if i % args.print_freq == 0:
print_log('Epoch: [{0}][{1}/{2}]\t'
'Time {batch_time.val:.3f} ({batch_time.avg:.3f})\t'
'Data {data_time.val:.3f} ({data_time.avg:.3f})\t'
'Loss {loss.val:.4f} ({loss.avg:.4f})\t'
'Prec@1 {top1.val:.3f} ({top1.avg:.3f})\t'
'Prec@5 {top5.val:.3f} ({top5.avg:.3f})'.format(
epoch, i, len(train_loader), batch_time=batch_time,
data_time=data_time, loss=losses, top1=top1, top5=top5), log)
def validate(val_loader, model, criterion, log):
batch_time = AverageMeter()
losses = AverageMeter()
top1 = AverageMeter()
top5 = AverageMeter()
# switch to evaluate mode
model.eval()
end = time.time()
for i, (input, target) in enumerate(val_loader):
target = target.cuda(async=True)
input_var = torch.autograd.Variable(input, volatile=True)
target_var = torch.autograd.Variable(target, volatile=True)
# compute output
output = model(input_var)
loss = criterion(output, target_var)
# measure accuracy and record loss
prec1, prec5 = accuracy(output.data, target, topk=(1, 5))
losses.update(loss.data[0], input.size(0))
top1.update(prec1[0], input.size(0))
top5.update(prec5[0], input.size(0))
# measure elapsed time
batch_time.update(time.time() - end)
end = time.time()
if i % args.print_freq == 0:
print_log('Test: [{0}/{1}]\t'
'Time {batch_time.val:.3f} ({batch_time.avg:.3f})\t'
'Loss {loss.val:.4f} ({loss.avg:.4f})\t'
'Prec@1 {top1.val:.3f} ({top1.avg:.3f})\t'
'Prec@5 {top5.val:.3f} ({top5.avg:.3f})'.format(
i, len(val_loader), batch_time=batch_time, loss=losses,
top1=top1, top5=top5), log)
print_log(' * Prec@1 {top1.avg:.3f} Prec@5 {top5.avg:.3f} Error@1 {error1:.3f}'.format(top1=top1, top5=top5,
error1=100 - top1.avg), log)
return top1.avg
def save_checkpoint(state, is_best, filename, bestname):
torch.save(state, filename)
if is_best:
shutil.copyfile(filename, bestname)
def print_log(print_string, log):
print("{:}".format(print_string))
log.write('{:}\n'.format(print_string))
log.flush()
class AverageMeter(object):
"""Computes and stores the average and current value"""
def __init__(self):
self.reset()
def reset(self):
self.val = 0
self.avg = 0
self.sum = 0
self.count = 0
def update(self, val, n=1):
self.val = val
self.sum += val * n
self.count += n
self.avg = self.sum / self.count
def adjust_learning_rate(optimizer, epoch):
"""Sets the learning rate to the initial LR decayed by 10 every 30 epochs"""
lr = args.lr * (0.1 ** (epoch // args.lr_adjust))
for param_group in optimizer.param_groups:
param_group['lr'] = lr
def accuracy(output, target, topk=(1,)):
"""Computes the precision@k for the specified values of k"""
maxk = max(topk)
batch_size = target.size(0)
_, pred = output.topk(maxk, 1, True, True)
pred = pred.t()
correct = pred.eq(target.view(1, -1).expand_as(pred))
res = []
for k in topk:
correct_k = correct[:k].view(-1).float().sum(0, keepdim=True)
res.append(correct_k.mul_(100.0 / batch_size))
return res
class Mask:
def __init__(self, model):
self.model_size = {}
self.model_length = {}
self.compress_rate = {}
self.distance_rate = {}
self.mat = {}
self.model = model
self.mask_index = []
self.filter_small_index = {}
self.filter_large_index = {}
self.similar_matrix = {}
def get_codebook(self, weight_torch, compress_rate, length):
weight_vec = weight_torch.view(length)
weight_np = weight_vec.cpu().numpy()
weight_abs = np.abs(weight_np)
weight_sort = np.sort(weight_abs)
threshold = weight_sort[int(length * (1 - compress_rate))]
weight_np[weight_np <= -threshold] = 1
weight_np[weight_np >= threshold] = 1
weight_np[weight_np != 1] = 0
print("codebook done")
return weight_np
def get_filter_codebook(self, weight_torch, compress_rate, length):
codebook = np.ones(length)
if len(weight_torch.size()) == 4:
filter_pruned_num = int(weight_torch.size()[0] * (1 - compress_rate))
weight_vec = weight_torch.view(weight_torch.size()[0], -1)
# norm1 = torch.norm(weight_vec, 1, 1)
# norm1_np = norm1.cpu().numpy()
norm2 = torch.norm(weight_vec, 2, 1)
norm2_np = norm2.cpu().numpy()
filter_index = norm2_np.argsort()[:filter_pruned_num]
# norm1_sort = np.sort(norm1_np)
# threshold = norm1_sort[int (weight_torch.size()[0] * (1-compress_rate) )]
kernel_length = weight_torch.size()[1] * weight_torch.size()[2] * weight_torch.size()[3]
for x in range(0, len(filter_index)):
codebook[filter_index[x] * kernel_length: (filter_index[x] + 1) * kernel_length] = 0
print("filter codebook done")
elif len(weight_torch.size()) == 2:
weight_torch = weight_torch.view(weight_torch.size()[0], weight_torch.size()[1], 1, 1)
codebook = self.get_filter_codebook(weight_torch, compress_rate, length)
print("filter codebook for fc done")
else:
pass
return codebook
@timing
def get_filter_similar_old(self, weight_torch, compress_rate, distance_rate, length):
codebook = np.ones(length)
if len(weight_torch.size()) == 4:
filter_pruned_num = int(weight_torch.size()[0] * (1 - compress_rate))
similar_pruned_num = int(weight_torch.size()[0] * distance_rate)
weight_vec = weight_torch.view(weight_torch.size()[0], -1)
# norm1 = torch.norm(weight_vec, 1, 1)
# norm1_np = norm1.cpu().numpy()
norm2 = torch.norm(weight_vec, 2, 1)
norm2_np = norm2.cpu().numpy()
filter_small_index = []
filter_large_index = []
filter_large_index = norm2_np.argsort()[filter_pruned_num:]
filter_small_index = norm2_np.argsort()[:filter_pruned_num]
print('weight_vec.size', weight_vec.size())
# distance using pytorch function
similar_matrix = torch.zeros((len(filter_large_index), len(filter_large_index)))
for x1, x2 in enumerate(filter_large_index):
for y1, y2 in enumerate(filter_large_index):
# cos = torch.nn.CosineSimilarity(dim=1, eps=1e-6)
# similar_matrix[x1, y1] = cos(weight_vec[x2].view(1, -1), weight_vec[y2].view(1, -1))[0]
pdist = torch.nn.PairwiseDistance(p=2)
# print('weight_vec[x2].size', weight_vec[x2].size())
similar_matrix[x1, y1] = pdist(weight_vec[x2].view(1, -1), weight_vec[y2].view(1, -1))[0][0]
# print('weight_vec[x2].size after', weight_vec[x2].size())
# more similar with other filter indicates large in the sum of row
similar_sum = torch.sum(torch.abs(similar_matrix), 0).numpy()
# for distance similar: get the filter index with largest similarity == small distance
similar_large_index = similar_sum.argsort()[similar_pruned_num:]
similar_small_index = similar_sum.argsort()[: similar_pruned_num]
similar_index_for_filter = [filter_large_index[i] for i in similar_small_index]
print('filter_large_index', filter_large_index)
print('filter_small_index', filter_small_index)
print('similar_sum', similar_sum)
print('similar_large_index', similar_large_index)
print('similar_small_index', similar_small_index)
print('similar_index_for_filter', similar_index_for_filter)
kernel_length = weight_torch.size()[1] * weight_torch.size()[2] * weight_torch.size()[3]
for x in range(0, len(similar_index_for_filter)):
codebook[
similar_index_for_filter[x] * kernel_length: (similar_index_for_filter[x] + 1) * kernel_length] = 0
print("similar index done")
else:
pass
return codebook
# optimize for fast ccalculation
def get_filter_similar(self, weight_torch, compress_rate, distance_rate, length):
codebook = np.ones(length)
if len(weight_torch.size()) == 4:
filter_pruned_num = int(weight_torch.size()[0] * (1 - compress_rate))
similar_pruned_num = int(weight_torch.size()[0] * distance_rate)
weight_vec = weight_torch.view(weight_torch.size()[0], -1)
# norm1 = torch.norm(weight_vec, 1, 1)
# norm1_np = norm1.cpu().numpy()
norm2 = torch.norm(weight_vec, 2, 1)
norm2_np = norm2.cpu().numpy()
filter_small_index = []
filter_large_index = []
filter_large_index = norm2_np.argsort()[filter_pruned_num:]
filter_small_index = norm2_np.argsort()[:filter_pruned_num]
# # distance using pytorch function
# similar_matrix = torch.zeros((len(filter_large_index), len(filter_large_index)))
# for x1, x2 in enumerate(filter_large_index):
# for y1, y2 in enumerate(filter_large_index):
# # cos = torch.nn.CosineSimilarity(dim=1, eps=1e-6)
# # similar_matrix[x1, y1] = cos(weight_vec[x2].view(1, -1), weight_vec[y2].view(1, -1))[0]
# pdist = torch.nn.PairwiseDistance(p=2)
# similar_matrix[x1, y1] = pdist(weight_vec[x2].view(1, -1), weight_vec[y2].view(1, -1))[0][0]
# # more similar with other filter indicates large in the sum of row
# similar_sum = torch.sum(torch.abs(similar_matrix), 0).numpy()
# distance using numpy function
indices = torch.LongTensor(filter_large_index).cuda()
weight_vec_after_norm = torch.index_select(weight_vec, 0, indices).cpu().numpy()
# for euclidean distance
similar_matrix = distance.cdist(weight_vec_after_norm, weight_vec_after_norm, 'euclidean')
# for cos similarity
# similar_matrix = 1 - distance.cdist(weight_vec_after_norm, weight_vec_after_norm, 'cosine')
similar_sum = np.sum(np.abs(similar_matrix), axis=0)
# for distance similar: get the filter index with largest similarity == small distance
similar_large_index = similar_sum.argsort()[similar_pruned_num:]
similar_small_index = similar_sum.argsort()[: similar_pruned_num]
similar_index_for_filter = [filter_large_index[i] for i in similar_small_index]
kernel_length = weight_torch.size()[1] * weight_torch.size()[2] * weight_torch.size()[3]
for x in range(0, len(similar_index_for_filter)):
codebook[
similar_index_for_filter[x] * kernel_length: (similar_index_for_filter[x] + 1) * kernel_length] = 0
print("similar index done")
else:
pass
return codebook
def convert2tensor(self, x):
x = torch.FloatTensor(x)
return x
def init_length(self):
for index, item in enumerate(self.model.parameters()):
self.model_size[index] = item.size()
for index1 in self.model_size:
for index2 in range(0, len(self.model_size[index1])):
if index2 == 0:
self.model_length[index1] = self.model_size[index1][0]
else:
self.model_length[index1] *= self.model_size[index1][index2]
def init_rate(self, rate_norm_per_layer, rate_dist_per_layer):
if 'vgg' in args.arch:
cfg_official = [64, 64, 128, 128, 256, 256, 256, 512, 512, 512, 512, 512, 512]
cfg_CP_5x = [24, 22, 41, 51, 108, 89, 111, 184, 276, 228, 512, 512, 512]
# cfg = [32, 64, 128, 128, 256, 256, 256, 256, 256, 256, 256, 256, 256]
cfg_Thinet_conv = [32, 32, 64, 64, 128, 128, 128, 256, 256, 256, 512, 512, 512]
if args.VGG_pruned_style == "CP_5x":
cfg_now = cfg_CP_5x
elif args.VGG_pruned_style == "Thinet_conv":
cfg_now = cfg_Thinet_conv
cfg_index = 0
previous_cfg = True
for index, item in enumerate(self.model.named_parameters()):
self.compress_rate[index] = 1
if len(item[1].size()) == 4:
if not previous_cfg:
self.compress_rate[index] = rate_norm_per_layer
self.distance_rate[index] = rate_dist_per_layer
self.mask_index.append(index)
print(item[0], "self.mask_index", self.mask_index)
else:
self.compress_rate[index] = 1
self.distance_rate[index] = 1 - cfg_now[cfg_index] / item[1].size()[0]
self.mask_index.append(index)
print(item[0], "self.mask_index", self.mask_index, cfg_index, cfg_now[cfg_index])
cfg_index += 1
elif "resnet" in args.arch:
for index, item in enumerate(self.model.parameters()):
self.compress_rate[index] = 1
self.distance_rate[index] = 1
for key in range(args.layer_begin, args.layer_end + 1, args.layer_inter):
self.compress_rate[key] = rate_norm_per_layer
self.distance_rate[key] = rate_dist_per_layer
# different setting for different architecture
if args.arch == 'resnet18':
# last index include last fc layer
last_index = 60
skip_list = [21, 36, 51]
elif args.arch == 'resnet34':
last_index = 108
skip_list = [27, 54, 93]
elif args.arch == 'resnet50':
last_index = 159
skip_list = [12, 42, 81, 138]
elif args.arch == 'resnet101':
last_index = 312
skip_list = [12, 42, 81, 291]
elif args.arch == 'resnet152':
last_index = 465
skip_list = [12, 42, 117, 444]
self.mask_index = [x for x in range(0, last_index, 3)]
# skip downsample layer
if args.skip_downsample == 1:
for x in skip_list:
self.compress_rate[x] = 1
self.mask_index.remove(x)
print(self.mask_index)
else:
pass
def init_mask(self, rate_norm_per_layer, rate_dist_per_layer):
self.init_rate(rate_norm_per_layer, rate_dist_per_layer)
for index, item in enumerate(self.model.parameters()):
if index in self.mask_index:
# mask for norm criterion
self.mat[index] = self.get_filter_codebook(item.data, self.compress_rate[index],
self.model_length[index])
self.mat[index] = self.convert2tensor(self.mat[index])
if args.use_cuda:
self.mat[index] = self.mat[index].cuda()
# mask for distance criterion
self.similar_matrix[index] = self.get_filter_similar(item.data, self.compress_rate[index],
self.distance_rate[index],
self.model_length[index])
self.similar_matrix[index] = self.convert2tensor(self.similar_matrix[index])
if args.use_cuda:
self.similar_matrix[index] = self.similar_matrix[index].cuda()
print("mask Ready")
def do_mask(self):
for index, item in enumerate(self.model.parameters()):
if index in self.mask_index:
a = item.data.view(self.model_length[index])
b = a * self.mat[index]
item.data = b.view(self.model_size[index])
print("mask Done")
def do_similar_mask(self):
for index, item in enumerate(self.model.parameters()):
if index in self.mask_index:
a = item.data.view(self.model_length[index])
b = a * self.similar_matrix[index]
item.data = b.view(self.model_size[index])
print("mask similar Done")
def do_grad_mask(self):
for index, item in enumerate(self.model.parameters()):
if index in self.mask_index:
a = item.grad.data.view(self.model_length[index])
# reverse the mask of model
# b = a * (1 - self.mat[index])
b = a * self.mat[index]
b = b * self.similar_matrix[index]
item.grad.data = b.view(self.model_size[index])
# print("grad zero Done")
def if_zero(self):
for index, item in enumerate(self.model.parameters()):
if index in self.mask_index:
# if index in [x for x in range(args.layer_begin, args.layer_end + 1, args.layer_inter)]:
a = item.data.view(self.model_length[index])
b = a.cpu().numpy()
print("layer: %d, number of nonzero weight is %d, zero is %d" % (
index, np.count_nonzero(b), len(b) - np.count_nonzero(b)))
if __name__ == '__main__':
main()