-
Notifications
You must be signed in to change notification settings - Fork 3
/
main.py
471 lines (413 loc) · 17.5 KB
/
main.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
# Copyright (c) Microsoft Corporation.
# Licensed under the MIT License.
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import os
import pickle
import shutil
import sys
import random
import json
import time
import datetime
import argparse
from itertools import chain
import numpy as np
import torch
from torch import nn
from torch import optim
from torch.utils.data import DataLoader
import torch.nn.functional as Func
from child_model.sst_dataset import read_data_sst, SSTDataset
from common.utils import train_ops, count_model_params, update_lr
from common.utils import Logger, print_user_flags
from common.utils import DEFINE_boolean, DEFINE_float, DEFINE_integer, DEFINE_string
from controller.rl_controller import RLController
from child_model.textnas_child import TextNASChild
parse = argparse.ArgumentParser()
DEFINE_boolean(parse, "reset_output_dir", False, "Delete output_dir if exists.")
DEFINE_string(parse, "embedding_model", "glove", "")
DEFINE_string(parse, "child_fixed_arc", None, "")
DEFINE_string(parse, "data_path", "", "")
DEFINE_string(parse, "embedding_path", "", "")
DEFINE_string(parse, "output_dir", "", "")
DEFINE_string(parse, "search_for", "macro", "Must be [macro]")
DEFINE_string(parse, "controller_type", "rl", "random")
DEFINE_string(parse, "child_lr_decay_scheme", "cosine", "Strategy to decay learning "
"rate, must be ['cosine', 'noam', 'exponential', 'auto']")
DEFINE_integer(parse, "batch_size", 128, "")
DEFINE_integer(parse, "eval_batch_size", 128, "")
DEFINE_integer(parse, "class_num", 5, "")
DEFINE_integer(parse, "max_input_length", 64, "")
DEFINE_integer(parse, "global_seed", 1234, "")
DEFINE_integer(parse, "num_epochs", 10, "")
DEFINE_integer(parse, "child_num_layers", 24, "")
DEFINE_integer(parse, "child_filter_size", 5, "")
DEFINE_integer(parse, "child_out_filters", 256, "")
DEFINE_integer(parse, "child_out_filters_scale", 1, "")
DEFINE_integer(parse, "child_num_branches", 8, "")
DEFINE_integer(parse, "child_progressive_branches", 4, "")
DEFINE_integer(parse, "child_lr_T_0", 10, "for lr schedule")
DEFINE_integer(parse, "child_lr_T_mul", 2, "for lr schedule")
DEFINE_integer(parse, "min_count", 1, "")
DEFINE_integer(parse, "num_last_layer_output", 0, "last n layers as output, 0 for all")
DEFINE_float(parse, "train_ratio", 1.0, "")
DEFINE_float(parse, "valid_ratio", 1.0, "")
DEFINE_float(parse, "child_grad_bound", 5.0, "Gradient clipping")
DEFINE_float(parse, "child_lr", 0.02, "")
DEFINE_float(parse, "cnn_keep_prob", 0.8, "")
DEFINE_float(parse, "final_output_keep_prob", 1.0, "")
DEFINE_float(parse, "lstm_out_keep_prob", 0.8, "")
DEFINE_float(parse, "embed_keep_prob", 0.8, "")
DEFINE_float(parse, "attention_keep_prob", 0.8, "")
DEFINE_float(parse, "child_l2_reg", 3e-6, "")
DEFINE_float(parse, "child_lr_max", 0.002, "for lr schedule")
DEFINE_float(parse, "child_lr_min", 0.001, "for lr schedule")
DEFINE_string(parse, "child_optim_algo", "adam", "")
DEFINE_string(parse, "output_type", "avg_pool", "")
DEFINE_boolean(parse, "multi_path", False, "Search for multiple path in the architecture")
DEFINE_boolean(parse, "is_binary", False, "binary label for sst dataset")
DEFINE_boolean(parse, "all_layer_output", True, "use all layer as output")
DEFINE_boolean(parse, "output_linear_combine", True, "")
DEFINE_boolean(parse, "is_mask", True, "whether apply mask")
DEFINE_boolean(parse, "fixed_seed", True, "")
DEFINE_boolean(parse, "load_checkpoint", False, "whether load checkpoint")
DEFINE_float(parse, "controller_lr", 1e-3, "")
DEFINE_float(parse, "controller_l2_reg", 1e-3, "")
DEFINE_float(parse, "controller_bl_dec", 0.99, "")
DEFINE_float(parse, "controller_tanh_constant", None, "")
DEFINE_float(parse, "controller_temperature", None, "")
DEFINE_float(parse, "controller_entropy_weight", None, "")
DEFINE_float(parse, "controller_skip_target", 0.8, "")
DEFINE_float(parse, "controller_skip_weight", 0.0, "")
DEFINE_integer(parse, "controller_train_steps", 500, "")
DEFINE_integer(parse, "controller_train_every", 2,
"train the controller after this number of epochs")
DEFINE_boolean(parse, "controller_training", False, "")
DEFINE_boolean(parse, "controller_use_critic", False, "")
DEFINE_boolean(parse, "is_cuda", True, "")
DEFINE_integer(parse, "log_every", 50, "How many steps to log")
DEFINE_integer(parse, "eval_every_epochs", 1, "How many epochs to eval")
FLAGS = parse.parse_args()
def set_random_seed(seed):
print("-" * 80)
print("set random seed for data reading: {}".format(seed))
random.seed(seed)
os.environ['PYTHONHASHSEED'] = str(seed)
np.random.seed(seed)
random.seed(seed)
torch.manual_seed(seed)
torch.cuda.manual_seed(seed)
torch.backends.cudnn.deterministic = True
def get_model(embedding, num_layers, pre_idxs=[]):
print("num layers: {0}".format(num_layers))
assert FLAGS.search_for is not None, "Please specify --search_for"
if FLAGS.search_for == "micro":
raise ValueError("Currently only 'macro' search supported.")
else:
if FLAGS.controller_type == "rl":
ControllerClass = RLController
else:
ControllerClass = None
child_model = TextNASChild(
embedding=embedding,
fixed_arc=FLAGS.child_fixed_arc,
num_layers=num_layers,
out_filters_scale=FLAGS.child_out_filters_scale,
out_filters=FLAGS.child_out_filters,
cnn_keep_prob=FLAGS.cnn_keep_prob,
final_output_keep_prob=FLAGS.final_output_keep_prob,
embed_keep_prob=FLAGS.embed_keep_prob,
lstm_out_keep_prob=FLAGS.lstm_out_keep_prob,
attention_keep_prob=FLAGS.attention_keep_prob,
multi_path=FLAGS.multi_path,
embedding_model=FLAGS.embedding_model,
all_layer_output=FLAGS.all_layer_output,
output_linear_combine=FLAGS.output_linear_combine,
num_last_layer_output=FLAGS.num_last_layer_output,
is_mask=FLAGS.is_mask,
output_type=FLAGS.output_type,
max_input_length=FLAGS.max_input_length,
class_num=FLAGS.class_num,
is_cuda=FLAGS.is_cuda)
if FLAGS.child_fixed_arc is None:
controller_model = ControllerClass(
search_for=FLAGS.search_for,
skip_target=FLAGS.controller_skip_target,
skip_weight=FLAGS.controller_skip_weight,
num_layers=num_layers,
num_branches=FLAGS.child_num_branches,
out_filters=FLAGS.child_out_filters,
lstm_size=64,
lstm_num_layers=1,
lstm_keep_prob=1.0,
tanh_constant=FLAGS.controller_tanh_constant,
temperature=FLAGS.controller_temperature,
lr_init=FLAGS.controller_lr,
l2_reg=FLAGS.controller_l2_reg,
entropy_weight=FLAGS.controller_entropy_weight,
bl_dec=FLAGS.controller_bl_dec,
use_critic=FLAGS.controller_use_critic,
optim_algo="adam",
pre_idxs=pre_idxs,
multi_path=FLAGS.multi_path,
is_cuda=FLAGS.is_cuda)
else:
controller_model = None
return child_model, controller_model
def print_arc(arc, num_layers):
start = 0
for i in range(0, num_layers):
end = start + i + 1
if FLAGS.multi_path:
end += 1
out_str = "fixed_arc=\"$fixed_arc {0}\"".format(np.reshape(arc[start: end], [-1]))
out_str = out_str.replace("[", "").replace("]", "")
print(out_str)
start = end
def eval_once(child_model, eval_set, criterion, valid_dataloader=None, test_dataloader=None):
if eval_set == "test":
assert test_dataloader is not None
dataloader = test_dataloader
elif eval_set == "valid":
assert valid_dataloader is not None
dataloader = valid_dataloader
else:
raise NotImplementedError("Unknown eval_set '{}'".format(eval_set))
tot_acc = 0
tot = 0
losses = []
with torch.no_grad(): # save memory
for batch in dataloader:
sent_ids, mask, labels = batch
if FLAGS.is_cuda:
sent_ids = sent_ids.cuda()
mask = mask.cuda()
labels = labels.cuda()
logits = child_model(sent_ids, mask) # run
loss = criterion(logits, labels.long())
loss = loss.mean()
preds = logits.argmax(dim=1).long()
acc = torch.eq(preds, labels.long()).long().sum().item()
losses.append(loss)
tot_acc += acc
tot += len(labels)
losses = torch.tensor(losses)
loss = losses.mean()
if tot > 0:
final_acc = float(tot_acc) / tot
else:
final_acc = 0
print("Error in calculating final_acc")
return final_acc, loss
def train(sents, mask, labels, output_dir, num_layers, embedding, pre_idxs=[]):
print("Build dataloader")
train_dataset = SSTDataset(sents["train"], mask["train"], labels["train"])
test_dataset = SSTDataset(sents["test"], mask["test"], labels["test"])
train_dataloader = DataLoader(train_dataset, batch_size=FLAGS.batch_size, shuffle=True, pin_memory=True)
test_dataloader = DataLoader(test_dataset, batch_size=FLAGS.eval_batch_size, pin_memory=True)
print("Build model")
print("-" * 80)
child_model, controller_model = get_model(embedding, num_layers, pre_idxs)
print("Finish build model")
for name, var in child_model.named_parameters():
print(name, var.size(), var.requires_grad) # output all params
num_vars = count_model_params(child_model.parameters())
print("Model has {} params".format(num_vars))
for m in child_model.modules(): # initializer
if isinstance(m, (nn.Conv1d, nn.Linear)):
nn.init.xavier_uniform_(m.weight)
criterion = nn.CrossEntropyLoss()
# get optimizer
if FLAGS.child_optim_algo == "adam":
optimizer = optim.Adam(child_model.parameters(), eps=1e-3, weight_decay=FLAGS.child_l2_reg) # with L2
elif FLAGS.child_optim_algo == "momentum":
optimizer = optim.SGD(self.parameters(), momentum=0.9, weight_decay=self.l2_reg, nesterov=True)
else:
raise ValueError("Unknown optim_algo {}".format(optim_algo))
if FLAGS.is_cuda:
child_model.cuda()
criterion.cuda()
if controller_model is not None:
controller_model.cuda()
if FLAGS.child_fixed_arc is not None:
fixed_arc = np.array([int(x) for x in FLAGS.child_fixed_arc.split(" ") if x])
print_arc(fixed_arc, num_layers)
print("Start training")
print("-" * 80)
start_time = time.time()
step = 0
# save path
model_save_path = os.path.join(FLAGS.output_dir, "model.pth")
best_model_save_path = os.path.join(FLAGS.output_dir, "best_model.pth")
best_acc = 0
start_epoch = 0
if FLAGS.load_checkpoint:
if os.path.isfile(model_save_path):
checkpoint = torch.load(model_save_path, map_location = torch.device('cpu'))
step = checkpoint['step']
start_epoch = checkpoint['epoch']
child_model.load_state_dict(checkpoint['child_model_state_dict'])
optimizer.load_state_dict(checkpoint['optimizer_state_dict'])
if controller_model is not None:
controller_model.load_state_dict(checkpoint['controller_model_state_dict'])
for epoch in range(start_epoch, FLAGS.num_epochs):
lr = update_lr(optimizer,
epoch,
lr_decay_scheme=FLAGS.child_lr_decay_scheme,
lr_max=FLAGS.child_lr_max,
lr_min=FLAGS.child_lr_min,
lr_T_0=FLAGS.child_lr_T_0,
lr_T_mul=FLAGS.child_lr_T_mul)
child_model.train()
if controller_model is not None:
controller_model.eval()
for batch in train_dataloader:
sent_ids, mask, labels = batch
if FLAGS.is_cuda:
sent_ids = sent_ids.cuda()
mask = mask.cuda()
labels = labels.cuda()
step += 1
if controller_model is not None:
controller_model._build_sampler(pre_idxs)
child_model.get_sample_arc(controller_model.sample_arc)
logits = child_model(sent_ids, mask) # run
loss = criterion(logits, labels.long())
loss = loss.mean()
preds = logits.argmax(dim=1).long()
acc = torch.eq(preds, labels.long()).long().sum().item()
gn = train_ops(
loss,
child_model.parameters(),
optimizer,
clip_mode="norm",
grad_bound=FLAGS.child_grad_bound)
if step % FLAGS.log_every == 0:
curr_time = time.time()
log_string = ""
log_string += "epoch={:<6d}".format(epoch)
log_string += "ch_step={:<6d}".format(step)
log_string += " loss={:<8.6f}".format(loss)
log_string += " lr={:<8.4f}".format(lr)
log_string += " |g|={:<8.4f}".format(gn)
log_string += " tr_acc={:<3d}/{:>3d}".format(acc, logits.size()[0])
log_string += " mins={:<10.2f}".format(float(curr_time - start_time) / 60)
print(log_string)
epoch += 1
save_state = {
'step' : step,
'epoch' : epoch,
'child_model_state_dict' : child_model.state_dict(),
'optimizer_state_dict' : optimizer.state_dict()}
torch.save(save_state, model_save_path)
child_model.eval()
if controller_model is None:
print("Epoch {}: Eval".format(epoch))
eval_acc, eval_loss = eval_once(child_model, "test", criterion, test_dataloader=test_dataloader)
print("ch_step={} {}_accuracy={:<6.4f} {}_loss={:<6.4f}".format(
step, "test", eval_acc, "test", eval_loss))
if eval_acc > best_acc:
best_acc = eval_acc
print("Save best model")
save_state = {
'step' : step,
'epoch' : epoch,
'child_model_state_dict' : child_model.state_dict(),
'optimizer_state_dict' : optimizer.state_dict()}
torch.save(save_state, best_model_save_path)
else:
if (FLAGS.controller_training and epoch % FLAGS.controller_train_every == 0):
print("Epoch {}: Training controller".format(epoch))
controller_model.train()
for ct_step in range(FLAGS.controller_train_steps):
controller_model._build_sampler(pre_idxs)
child_model.get_sample_arc(controller_model.sample_arc)
eval_acc, eval_loss = eval_once(child_model, "test", criterion, test_dataloader=test_dataloader)
loss, entropy, lr, gn, reward, sample_log_prob, bl = controller_model.trainer(eval_acc, ct_step)
if ct_step % FLAGS.log_every == 0:
curr_time = time.time()
log_string = ""
log_string += "ctrl_step={:<6d}".format(ct_step)
log_string += " loss={:<7.3f}".format(loss)
log_string += " sample_log_prob={:<8.4f}".format(sample_log_prob)
log_string += " reward={:<8.4f}".format(reward)
log_string += " ent={:<5.2f}".format(entropy)
log_string += " lr={:<6.4f}".format(lr)
log_string += " |g|={:<8.4f}".format(gn)
log_string += " acc={:<6.4f}".format(eval_acc)
log_string += " bl={:<5.2f}".format(bl)
log_string += " mins={:<.2f}".format(float(curr_time - start_time) / 60)
print(log_string)
avg_acc = 0
max_acc = 0
print("Here are 10 architectures")
for _ in range(10):
controller_model._build_sampler(pre_idxs)
child_model.get_sample_arc(controller_model.sample_arc)
arc = child_model.sample_arc
eval_acc, eval_loss = eval_once(child_model, "test", criterion, test_dataloader=test_dataloader)
print("ch_step={} {}_accuracy={:<6.4f} {}_loss={:<6.4f}".format(
step, "test", eval_acc, "test", eval_loss))
if FLAGS.search_for == "micro":
normal_arc, reduce_arc = arc
print(np.reshape(normal_arc, [-1]))
print(np.reshape(reduce_arc, [-1]))
else:
start = 0
for layer_id in range(num_layers):
end = start + 1 + layer_id
if FLAGS.multi_path:
end += 1
out_str = "fixed_arc=\"$fixed_arc {0}\"".format(np.reshape(arc[start: end], [-1]))
out_str = out_str.replace("[", "").replace("]", "")
print(out_str)
start = end
if eval_acc > max_acc:
max_acc = eval_acc
avg_acc += eval_acc
print("-" * 80)
avg_acc /= 10.0
threshold = max_acc
print("sample_arc_acc, {0}, {1}".format(avg_acc, max_acc))
if max_acc > best_acc:
best_acc = max_acc
print("Found better model at: {}".format(step))
save_state = {
'step' : step,
'epoch' : epoch,
'child_model_state_dict' : child_model.state_dict(),
'controller_model_state_dict' : controller_model.state_dict(),
'optimizer_state_dict' : optimizer.state_dict()}
torch.save(save_state, best_model_save_path)
return eval_acc
def main():
print("-" * 80)
if not os.path.isdir(FLAGS.output_dir):
print("Path {} does not exist. Creating.".format(FLAGS.output_dir))
os.makedirs(FLAGS.output_dir)
elif FLAGS.reset_output_dir:
print("Path {} exists. Remove and remake.".format(FLAGS.output_dir))
shutil.rmtree(FLAGS.output_dir, ignore_errors=True)
os.makedirs(FLAGS.output_dir)
print("-" * 80)
log_file = os.path.join(FLAGS.output_dir, "stdout")
print("Logging to {}".format(log_file))
sys.stdout = Logger(log_file)
print_user_flags(FLAGS)
if FLAGS.fixed_seed:
set_random_seed(FLAGS.global_seed)
print("load data")
sents, mask, labels, embedding = read_data_sst(FLAGS.data_path,
FLAGS.max_input_length,
FLAGS.embedding_model,
FLAGS.min_count,
FLAGS.train_ratio,
FLAGS.valid_ratio,
FLAGS.embedding_path,
FLAGS.is_binary)
print("load data finish")
train(sents, mask, labels, FLAGS.output_dir, FLAGS.child_num_layers, embedding)
if __name__ == "__main__":
main()