-
Notifications
You must be signed in to change notification settings - Fork 0
/
run.py
956 lines (721 loc) · 33 KB
/
run.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
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
# -*- coding: utf-8 -*-
'''
Build and train the query reformulator model
'''
import cPickle as pkl
import time
import numpy as np
import theano
import theano.tensor as tensor
import utils
import corpus_hdf5
import dataset_hdf5
import parameters as prm
import nltk
import random
import sys
import h5py
from theano import config
from theano.sandbox.rng_mrg import MRG_RandomStreams as RandomStreams
from collections import OrderedDict
from sklearn.decomposition import PCA
from op_search import Search
reload(sys)
sys.setdefaultencoding('utf8')
# only print four decimals on float arrays.
np.set_printoptions(linewidth=150, formatter={'float': lambda x: "{0:0.4f}".format(x)})
# Set the random number generators' seeds for consistency
SEED = 123
np.random.seed(SEED)
def numpy_floatX(data):
return np.asarray(data, dtype=config.floatX)
def np_floatX(data):
return np.asarray(data, dtype=config.floatX)
def _slice(_x, n, dim):
if _x.ndim == 3:
return _x[:, :, n * dim:(n + 1) * dim]
return _x[:, n * dim:(n + 1) * dim]
def get_minibatches_idx(n, minibatch_size, shuffle=False, max_samples=None):
"""
Used to shuffle the dataset at each iteration.
"""
idx_list = np.arange(n, dtype="int32")
if shuffle:
np.random.shuffle(idx_list)
if max_samples:
idx_list = idx_list[:max_samples]
n = max_samples
minibatches = []
minibatch_start = 0
for i in range(n // minibatch_size):
minibatches.append(idx_list[minibatch_start:
minibatch_start + minibatch_size])
minibatch_start += minibatch_size
if (minibatch_start != n):
# Make a minibatch out of what is left
minibatches.append(idx_list[minibatch_start:])
return zip(range(len(minibatches)), minibatches)
def zipp(params, tparams):
"""
When we reload the model. Needed for the GPU stuff.
"""
for kk, vv in params.iteritems():
tparams[kk].set_value(vv)
def unzip(zipped):
"""
When we pickle the model. Needed for the GPU stuff.
"""
new_params = OrderedDict()
for kk, vv in zipped.iteritems():
new_params[kk] = vv.get_value()
return new_params
def dropout_layer(state_before, is_train, trng):
proj = tensor.switch(is_train,
(state_before *
trng.binomial(state_before.shape,
p=(1-prm.dropout), n=1,
dtype=state_before.dtype)),
state_before * (1-prm.dropout))
return proj
def load_params(path, params):
pp = np.load(path)
for kk, vv in params.iteritems():
if kk in pp:
if params[kk].shape == pp[kk].shape:
params[kk] = pp[kk]
else:
print 'The shape of layer', kk, params[kk].shape, 'is different from shape of the stored layer with the same name', pp[kk].shape, '.'
else:
print '%s is not in the archive' % kk
return params
def load_wemb(params, vocab):
wemb = pkl.load(open(prm.wordemb_path, 'rb'))
dim_emb_orig = wemb.values()[0].shape[0]
W = 0.01 * np.random.randn(prm.n_words, dim_emb_orig).astype(config.floatX)
for word, pos in vocab.items():
if word in wemb:
W[pos,:] = wemb[word]
if prm.dim_emb < dim_emb_orig:
pca =PCA(n_components=prm.dim_emb, copy=False, whiten=True)
W = pca.fit_transform(W)
params['W'] = W
return params
def itemlist(tparams):
return [vv for kk, vv in tparams.iteritems()]
def init_tparams(params):
tparams = OrderedDict()
for kk, pp in params.iteritems():
tparams[kk] = theano.shared(params[kk], name=kk)
return tparams
def ortho_weight(ndim):
W = np.random.randn(ndim, ndim)
u, s, v = np.linalg.svd(W)
return u.astype(config.floatX)
def matrix(dim):
return np.concatenate([ortho_weight(dim),
ortho_weight(dim),
ortho_weight(dim),
ortho_weight(dim)], axis=1)
def softmax_mask(x, mask=None):
m = tensor.max(x, axis=-1, keepdims=True)
if mask:
e_x = tensor.exp(x - m) * mask
else:
e_x = tensor.exp(x - m)
return e_x / tensor.maximum(e_x.sum(axis=-1, keepdims=True), 1e-8) #this small constant avoids possible division by zero created by the mask
def lstm_layer(x, h_, c_):
i = tensor.nnet.sigmoid(_slice(x, 0, prm.dim_proj))
f = tensor.nnet.sigmoid(_slice(x, 1, prm.dim_proj))
o = tensor.nnet.sigmoid(_slice(x, 2, prm.dim_proj))
c = tensor.tanh(_slice(x, 3, prm.dim_proj))
c = f * c_ + i * c
h = o * tensor.tanh(c)
return h, c
def init_params(options):
params = OrderedDict()
exclude_params = {}
params['W'] = 0.01 * np.random.randn(prm.n_words, prm.dim_emb).astype(config.floatX) # vocab to word embeddings
params['UNK'] = 0.01 * np.random.randn(1, prm.dim_emb).astype(config.floatX) # vector for unknown words.
n_features = [prm.dim_emb,] + prm.filters_query
for i in range(len(prm.filters_query)):
params['Ww_att_q'+str(i)] = 0.01 * np.random.randn(n_features[i+1], n_features[i], 1, prm.window_query[i]).astype(config.floatX)
params['bw_att_q'+str(i)] = np.zeros((n_features[i+1],)).astype(config.floatX) # bias score
params['Aq'] = 0.01 * np.random.randn(n_features[-1], prm.dim_proj).astype(config.floatX) # score
n_hidden_actor = [prm.dim_proj] + prm.n_hidden_actor + [2]
for i in range(len(n_hidden_actor)-1):
params['V'+str(i)] = 0.01 * np.random.randn(n_hidden_actor[i], n_hidden_actor[i+1]).astype(config.floatX) # score
params['bV'+str(i)] = np.zeros((n_hidden_actor[i+1],)).astype(config.floatX) # bias score
# set initial bias towards not selecting words.
params['bV'+str(i)] = np.array([10., 0.]).astype(config.floatX) # bias score
n_hidden_critic = [prm.dim_proj] + prm.n_hidden_critic + [1]
for i in range(len(n_hidden_critic)-1):
params['C'+str(i)] = 0.01 * np.random.randn(n_hidden_critic[i], n_hidden_critic[i+1]).astype(config.floatX) # score
params['bC'+str(i)] = np.zeros((n_hidden_critic[i+1],)).astype(config.floatX) # bias score
n_features = [prm.dim_emb,] + prm.filters_cand
for i in range(len(prm.filters_cand)):
params['Ww_att_c_0_'+str(i)] = 0.01 * np.random.randn(n_features[i+1], n_features[i], 1, prm.window_cand[i]).astype(config.floatX)
params['bw_att_c_0_'+str(i)] = np.zeros((n_features[i+1],)).astype(config.floatX) # bias score
params['Ad'] = 0.01 * np.random.randn(n_features[-1], prm.dim_proj).astype(config.floatX) # score
params['bAd'] = np.zeros((prm.dim_proj,)).astype(config.floatX) # bias score
if prm.fixed_wemb:
exclude_params['W'] = True
return params, exclude_params
def conv_query(q_a, tparams):
q_aw = q_a.dimshuffle(0, 2, 'x', 1)
for j in range(len(prm.filters_query)):
q_aw = tensor.nnet.conv2d(q_aw,
tparams['Ww_att_q'+str(j)],
border_mode=(0, prm.window_query[j]//2))
q_aw += tparams['bw_att_q'+str(j)][None,:,None,None]
q_aw = tensor.maximum(q_aw, 0.)
q_aw = q_aw[:, :, 0, :].dimshuffle(0, 2, 1)
q_a = q_aw.reshape((q_a.shape[0], q_a.shape[1], -1))
return q_a
def conv_cand(D_a, tparams, n_iter):
D_aw = D_a.reshape((-1, D_a.shape[2], D_a.shape[3]))
D_aw = D_aw.dimshuffle(0, 2, 'x', 1)
for j in range(len(prm.filters_cand)):
D_aw = tensor.nnet.conv2d(D_aw,
tparams['Ww_att_c_' + str(n_iter) + '_' + str(j)],
border_mode=(0, prm.window_cand[j]//2))
D_aw += tparams['bw_att_c_' + str(n_iter) + '_' + str(j)][None,:,None,None]
D_aw = tensor.maximum(D_aw, 0.)
D_aw = D_aw[:, :, 0, :].dimshuffle(0, 2, 1)
D_a = D_aw.reshape((D_a.shape[0], D_a.shape[1], D_a.shape[2], -1))
return D_a
def f(q_i, D_gt_id, tparams, is_train, trng, options):
# Use search engine again to compute the reward/metrics given a query.
search = Search(options)
# append the unknown vector for words whose index = -1.
W_ = tensor.concatenate([tparams['W'], tparams['UNK']], axis=0)
q_m = (q_i > -2).astype('float32')
#get embeddings for the queries
q_a = W_[q_i.flatten()].reshape((q_i.shape[0], q_i.shape[1], prm.dim_emb)) * q_m[:,:,None]
if len(prm.filters_query) > 0:
q_aa = conv_query(q_a, tparams)
else:
q_aa = q_a
q_a_avg = q_a.sum(1) / tensor.maximum(1., q_m.sum(1, keepdims=True))
out = []
for n_iter in range(prm.n_iterations):
if n_iter == 0 and prm.q_0_fixed_until >= prm.n_iterations:
prob = tensor.zeros((q_a.shape[0], prm.max_words_input, 2))
bl = tensor.zeros((q_a.shape[0],))
D_m_r = tensor.zeros((q_a.shape[0], prm.max_words_input))
else:
if n_iter > 0:
D_m_ = (D_i_ > -2).astype('float32')
D_a_ = W_[D_i_.flatten()].reshape((D_i_.shape[0], D_i_.shape[1], D_i_.shape[2], prm.dim_emb)) * D_m_[:,:,:,None]
else:
D_a_ = 1. * q_a[:,None,:,:]
D_m_ = 1. * q_m[:,None,:]
if len(prm.filters_cand) > 0:
D_aa_ = conv_cand(D_a_, tparams, 0)
else:
D_aa_ = D_a_
D_aa_ = tensor.dot(D_aa_, tparams['Ad']) + tparams['bAd']
if n_iter > 0:
if prm.q_0_fixed_until < 2:
D_a = tensor.concatenate([D_a, D_a_], axis=1)
D_aa = tensor.concatenate([D_aa, D_aa_], axis=1)
D_m = tensor.concatenate([D_m, D_m_], axis=1)
else:
D_a = D_a_
D_aa = D_aa_
D_m = D_m_
else:
D_a = D_a_
D_aa = D_aa_
D_m = D_m_
D_a_r = D_a.reshape((D_a.shape[0], -1, D_a.shape[3]))
D_aa_r = D_aa.reshape((D_aa.shape[0], -1, D_aa.shape[3]))
D_m_r = D_m.reshape((D_m.shape[0],-1))
q_aa_avg = q_aa.sum(1) / tensor.maximum(1., q_m.sum(1, keepdims=True))
q_aa_att = q_aa_avg[:,None,:]
q_aa_att = tensor.dot(q_aa_att, tparams['Aq'])
z = D_aa_r + q_aa_att
# estimate reward based on the query.
bl = theano.gradient.grad_scale(z, 0.1)
D_m_r_c = theano.gradient.disconnected_grad(D_m_r)
bl = bl.sum(1) / tensor.maximum(1., D_m_r_c.sum(1))[:,None]
for i in range(len(prm.n_hidden_critic)+1):
if prm.dropout > 0:
bl = dropout_layer(bl, is_train, trng)
bl = tensor.maximum(0., bl)
bl = tensor.dot(bl, tparams['C'+str(i)]) + tparams['bC'+str(i)]
bl = tensor.tanh(bl)
bl = bl.flatten()
for i in range(len(prm.n_hidden_actor)+1):
if prm.dropout > 0:
z = dropout_layer(z, is_train, trng)
z = tensor.maximum(0., z)
z = tensor.dot(z, tparams['V'+str(i)]) + tparams['bV'+str(i)]
prob = softmax_mask(z) * D_m_r[:,:,None]
# if training, sample. Otherwise, pick maximum probability.
s = trng.multinomial(n=1, pvals=prob.reshape((-1, 2)), dtype=prob.dtype)
s = s.reshape((prob.shape[0],prob.shape[1],prob.shape[2]))
#if frozen is enabled and this iteration is within its limit, pick maximum probability.
if prm.frozen_until > 0:
if n_iter < prm.frozen_until:
s = prob
res = tensor.eq(is_train,1.) * s + tensor.eq(is_train,0.) * prob
# final answer & valid words
ans = res.argmax(2) * D_m_r
if n_iter < prm.q_0_fixed_until:
ones = tensor.ones((q_a.shape[0], prm.max_words_input))
if n_iter > 0:
# select everything from the original query in the first iteration.
ans = tensor.concatenate([ones, ans], axis=1)
else:
ans = ones
metrics, D_i_, D_id_, D_gt_m_ = search(ans, D_gt_id, n_iter, is_train)
out.append([prob, ans, metrics, bl, D_m_r, D_id_])
return out
def sgd(lr, tparams, grads, iin, out, updates):
""" Stochastic Gradient Descent
:note: A more complicated version of sgd then needed. This is
done like that for adadelta and rmsprop.
"""
# New set of shared variable that will contain the gradient
# for a mini-batch.
gshared = [theano.shared(p.get_value() * 0., name='%s_grad' % k)
for k, p in tparams.items()]
gsup = [(gs, g) for gs, g in zip(gshared, grads)]
# Function that computes gradients for a mini-batch, but do not
# updates the weights.
f_grad_shared = theano.function(iin, out, updates=gsup + updates,
name='sgd_f_grad_shared')
pup = [(p, p - lr * g) for p, g in zip(tparams.values(), gshared)]
# Function that updates the weights from the previously computed
# gradient.
f_update = theano.function([lr], [], updates=pup,
name='sgd_f_update')
return f_grad_shared, f_update
def adadelta(lr, tparams, grads, iin, out, updates):
"""
An adaptive learning rate optimizer
Notes
-----
For more information, see [ADADELTA]_.
.. [ADADELTA] Matthew D. Zeiler, *ADADELTA: An Adaptive Learning
Rate Method*, arXiv:1212.5701.
"""
zipped_grads = [theano.shared(p.get_value() * numpy_floatX(0.),
name='%s_grad' % k)
for k, p in tparams.items()]
running_up2 = [theano.shared(p.get_value() * numpy_floatX(0.),
name='%s_rup2' % k)
for k, p in tparams.items()]
running_grads2 = [theano.shared(p.get_value() * numpy_floatX(0.),
name='%s_rgrad2' % k)
for k, p in tparams.items()]
zgup = [(zg, g) for zg, g in zip(zipped_grads, grads)]
rg2up = [(rg2, 0.95 * rg2 + 0.05 * (g ** 2))
for rg2, g in zip(running_grads2, grads)]
f_grad_shared = theano.function(iin, out, updates=zgup + rg2up + updates,
name='adadelta_f_grad_shared')
updir = [-tensor.sqrt(ru2 + 1e-6) / tensor.sqrt(rg2 + 1e-6) * zg
for zg, ru2, rg2 in zip(zipped_grads,
running_up2,
running_grads2)]
ru2up = [(ru2, 0.95 * ru2 + 0.05 * (ud ** 2))
for ru2, ud in zip(running_up2, updir)]
param_up = [(p, p + ud) for p, ud in zip(tparams.values(), updir)]
f_update = theano.function([lr], [], updates=ru2up + param_up,
on_unused_input='ignore',
name='adadelta_f_update')
return f_grad_shared, f_update
def rmsprop(lr, tparams, grads, iin, out, updates):
"""
A variant of SGD that scales the step size by running average of the
recent step norms.
Notes
-----
For more information, see [Hint2014]_.
.. [Hint2014] Geoff Hinton, *Neural Networks for Machine Learning*,
lecture 6a,
http://cs.toronto.edu/~tijmen/csc321/slides/lecture_slides_lec6.pdf
"""
zipped_grads = [theano.shared(p.get_value() * numpy_floatX(0.),
name='%s_grad' % k)
for k, p in tparams.items()]
running_grads = [theano.shared(p.get_value() * numpy_floatX(0.),
name='%s_rgrad' % k)
for k, p in tparams.items()]
running_grads2 = [theano.shared(p.get_value() * numpy_floatX(0.),
name='%s_rgrad2' % k)
for k, p in tparams.items()]
zgup = [(zg, g) for zg, g in zip(zipped_grads, grads)]
rgup = [(rg, 0.95 * rg + 0.05 * g) for rg, g in zip(running_grads, grads)]
rg2up = [(rg2, 0.95 * rg2 + 0.05 * (g ** 2))
for rg2, g in zip(running_grads2, grads)]
f_grad_shared = theano.function(iin, out,
updates=zgup + rgup + rg2up + updates,
name='rmsprop_f_grad_shared')
updir = [theano.shared(p.get_value() * numpy_floatX(0.),
name='%s_updir' % k)
for k, p in tparams.items()]
updir_new = [(ud, 0.9 * ud - 1e-4 * zg / tensor.sqrt(rg2 - rg ** 2 + 1e-4))
for ud, zg, rg, rg2 in zip(updir, zipped_grads, running_grads,
running_grads2)]
param_up = [(p, p + udn[1])
for p, udn in zip(tparams.values(), updir_new)]
f_update = theano.function([lr], [], updates=updir_new + param_up,
on_unused_input='ignore',
name='rmsprop_f_update')
return f_grad_shared, f_update
def adam(lr0, tparams, grads, iin, out, updates):
gshared = [theano.shared(p.get_value() * 0., name='%s_grad'%k) for k, p in tparams.iteritems()]
gsup = [(gs, g) for gs, g in zip(gshared, grads)]
f_grad_shared = theano.function(iin, out, updates=gsup+updates, \
on_unused_input='ignore', allow_input_downcast=True)
b1 = 0.1
b2 = 0.001
e = 1e-8
updates = []
i = theano.shared(np.float32(0.))
i_t = i + 1.
fix1 = 1. - b1**(i_t)
fix2 = 1. - b2**(i_t)
lr_t = lr0 * (tensor.sqrt(fix2) / fix1)
for p, g in zip(tparams.values(), gshared):
m = theano.shared(p.get_value() * 0.)
v = theano.shared(p.get_value() * 0.)
m_t = (b1 * g) + ((1. - b1) * m)
v_t = (b2 * tensor.sqr(g)) + ((1. - b2) * v)
g_t = m_t / (tensor.sqrt(v_t) + e)
p_t = p - (lr_t * g_t)
updates.append((m, m_t))
updates.append((v, v_t))
updates.append((p, p_t))
updates.append((i, i_t))
f_update = theano.function([lr0], lr_t, updates=updates, on_unused_input='ignore')
return f_grad_shared, f_update
def build_model(tparams, options):
trng = RandomStreams(SEED)
off = 1e-8 # small constant to avoid log 0 = -inf
consider_constant = []
is_train = tensor.fscalar('is_train') # if = 1, training time.
q_i = tensor.imatrix('q_i') # input query
D_gt_id = tensor.imatrix('D_gt_id')
out = f(q_i, D_gt_id, tparams, is_train, trng, options)
cost = 0.
out_p = []
out_s = []
reward_last = 0
for i, (prob, ans, metrics, bl, D_m_r, D_id) in enumerate(out):
learn = True
# if frozen until is enabled.
if prm.frozen_until > 0:
# do not learn if this iteration is less than frozen_until.
if i < prm.frozen_until:
learn = False
reward = metrics[:,prm.metrics_map[prm.reward.upper()]]
r = reward - reward_last - bl
# cost for the baseline
cost_bl = (r ** 2).sum()
if learn:
if i < prm.q_0_fixed_until:
cap = prm.max_words_input
else:
cap = 0
r_ = theano.gradient.disconnected_grad(r)
cost_i = r_[:,None] * (-tensor.log(prob[:,:,1] + off)) * ans[:,cap:]
cost += cost_i.sum()
cost += cost_bl
# entropy regularization
if prm.erate > 0.:
cost_ent = prm.erate * ((D_m_r[:,:,None] * prob * tensor.log(prob + off)).sum(axis=(1,2))).sum()
cost += cost_ent
else:
cost_ent = 0. * cost
reward_last = reward
out_p.extend([ans, metrics, D_id])
out_s.extend([prob, ans, metrics, bl, cost_bl, D_id])
if prm.l2reg > 0.:
cost_l2reg = 0.
for name, w in tparams.items():
#do not include bias.
if (not name.lower().startswith('b')) and (name.lower() != 'w'):
cost_l2reg += prm.l2reg * (w**2).sum()
cost += cost_l2reg
f_pred = theano.function([q_i, D_gt_id, is_train], out_p, updates=[], name='f_pred', on_unused_input='ignore')
iin = [q_i, D_gt_id, is_train]
out = [cost, cost_ent] + out_s
updates = []
return iin, out, updates, f_pred, consider_constant
def lst2matrix(lst):
maxdim = len(max(lst,key=len))
out = -np.ones((len(lst), maxdim), dtype=np.int32)
for i,item in enumerate(lst):
out[i,:min(len(item),maxdim)] = item[:maxdim]
return out
def get_samples(input_queries, target_docs, index, options):
qi = [utils.clean(input_queries[t].lower()) for t in index]
D_gt_title = [target_docs[t] for t in index]
D_gt_id_lst = []
for j, t in enumerate(index):
D_gt_id_lst.append([])
for title in D_gt_title[j]:
if title in options['engine'].title_id_map:
D_gt_id_lst[-1].append(options['engine'].title_id_map[title])
else:
print 'ground-truth doc not in index:', title
D_gt_id = lst2matrix(D_gt_id_lst)
qi_i, qi_lst_ = utils.text2idx2(qi, options['vocab'], prm.max_words_input)
qi_lst = []
for qii_lst in qi_lst_:
# append empty strings, so the list size becomes <dim>.
qi_lst.append(qii_lst + max(0, prm.max_words_input - len(qii_lst)) * [''])
return qi, qi_i, qi_lst, D_gt_id, D_gt_title
def pred_error(f_pred, input_queries, target_docs, options, iterator):
"""
Evaluate model on the metrics.
f_pred: Theano function computing the prediction
"""
n = 0.
metrics = np.zeros((prm.n_iterations,len(prm.metrics_map)), dtype=np.float32)
i = 0.
for _, index in iterator:
qi, qi_i, qi_lst, D_gt_id, D_gt_url = get_samples(input_queries, target_docs, index, options)
# share the current queries with the search engine.
options['current_queries'] = qi_lst
is_train = 0.
out = f_pred(qi_i, D_gt_id, is_train)
if i % prm.dispFreq == 0:
print '=================================================================='
print
print 'Input Query: ', qi[0].replace('\n','\\n')
print
print 'Target Docs: ', str(D_gt_url[0])
print
for j in range(prm.n_iterations):
ans = out.pop(0)
metrics_i = out.pop(0)
D_id = out.pop(0)
metrics[j] += metrics_i.sum(0)
if i % prm.dispFreq == 0:
print
print 'Iteration', j
print
print 'Retrieved Docs: ', str([options['engine'].id_title_map[d_id] for d_id in D_id[0]])
print
print 'Reformulated Query:', options['reformulated_queries'][j][0]
print
print 'Query ANS: ',
for kk, word in enumerate(options['current_queries'][0][:ans.shape[1]]):
if word not in options['vocab'] and word != '':
word += '<unk>'
if ans[0,kk] == 1:
word = word.upper()
print str(word),
print
if i % prm.dispFreq == 0:
print '=================================================================='
n += len(index)
i += 1.
metrics /= n
return metrics
def train():
if prm.optimizer.lower() == 'adam':
optimizer=adam
elif prm.optimizer.lower() == 'sgd':
optimizer=sgd
elif prm.optimizer.lower() == 'rmsprop':
optimizer=rmsprop
elif prm.optimizer.lower() == 'adadelta':
optimizer=adadelta
options = locals().copy()
print 'parameters:', str(options)
prm_k = vars(prm).keys()
prm_d = vars(prm)
prm_k.sort()
for x in prm_k:
if not x.startswith('__'):
print x,'=', prm_d[x]
print 'loading Vocabulary...'
vocab = utils.load_vocab(prm.vocab_path, prm.n_words)
options['vocab'] = vocab
options['vocabinv'] = {}
for k,v in vocab.items():
options['vocabinv'][v] = k
print 'Loading Environment...'
if prm.engine.lower() == 'lucene':
import lucene_search
options['engine'] = lucene_search.LuceneSearch()
elif prm.engine.lower() == 'elastic':
import elastic_search
options['engine'] = elastic_search.ElasticSearch()
print 'Loading Dataset...'
dh5 = dataset_hdf5.DatasetHDF5(prm.dataset_path)
qi_train = dh5.get_queries(dset='train')
dt_train = dh5.get_doc_ids(dset='train')
qi_valid = dh5.get_queries(dset='valid')
dt_valid = dh5.get_doc_ids(dset='valid')
qi_test = dh5.get_queries(dset='test')
dt_test = dh5.get_doc_ids(dset='test')
if prm.train_size == -1:
train_size = len(qi_train)
else:
train_size = min(prm.train_size, len(qi_train))
if prm.valid_size == -1:
valid_size = len(qi_valid)
else:
valid_size = min(prm.valid_size, len(qi_valid))
if prm.test_size == -1:
test_size = len(qi_test)
else:
test_size = min(prm.test_size, len(qi_test))
print '%d train examples' % len(qi_train)
print '%d valid examples' % len(qi_valid)
print '%d test examples' % len(qi_test)
# This create the initial parameters as np ndarrays.
# Dict name (string) -> np ndarray
params, exclude_params = init_params(options)
if prm.wordemb_path:
print 'loading pre-trained word embeddings'
params = load_wemb(params, vocab)
options['W'] = params['W']
if prm.reload_model:
load_params(prm.reload_model, params)
print 'Building model'
# This create Theano Shared Variable from the parameters.
# Dict name (string) -> Theano Tensor Shared Variable
# params and tparams have different copy of the weights.
tparams = init_tparams(params)
for kk, value in tparams.iteritems():
tparams[kk] = theano.shared(value, name=kk)
iin, out, updates, f_pred, consider_constant \
= build_model(tparams, options)
#get only parameters that are not in the exclude_params list
tparams_ = OrderedDict([(kk, vv) for kk, vv in tparams.iteritems() if kk not in exclude_params])
grads = tensor.grad(out[0], wrt=itemlist(tparams_), consider_constant=consider_constant)
lr = tensor.scalar(name='lr')
f_grad_shared, f_update = optimizer(lr, tparams_, grads, iin, out, updates)
history_errs = []
best_p = None
if prm.validFreq == -1:
validFreq = len(qi_train) / prm.batch_size_train
else:
validFreq = prm.validFreq
if prm.saveFreq == -1:
saveFreq = len(qi_train) / prm.batch_size_train
else:
saveFreq = prm.saveFreq
uidx = 0 # the number of update done
estop = False # early stop
start_time = time.time()
print 'Optimization'
try:
for eidx in xrange(prm.max_epochs):
n_samples = 0
# Get new shuffled index for the training set.
kf = get_minibatches_idx(len(qi_train), prm.batch_size_train, shuffle=True)
for _, train_index in kf:
st = time.time()
uidx += 1
qi, qi_i, qi_lst, D_gt_id, D_gt_url = get_samples(qi_train, dt_train, train_index, options)
# share the current queries with the search engine.
options['current_queries'] = qi_lst
n_samples += len(qi)
is_train = 1.
out = f_grad_shared(qi_i, D_gt_id, is_train)
cost = out.pop(0)
cost_ent = out.pop(0)
lr_t = f_update(prm.lrate)
if np.isnan(cost) or np.isinf(cost):
print 'NaN detected'
return 1., 1., 1.
if np.mod(uidx, prm.dispFreq) == 0:
print '\n================================================================================'
print 'Epoch', eidx, 'Update', uidx, 'Cost', cost, 'LR_t', lr_t
print 'Time Minibatch Update: ' + str(time.time() - st)
print 'Input Query: ', qi[0].replace('\n','\\n')
print
print 'Target Docs: ', str(D_gt_url[0])
print
print 'Input Query Vocab: ', utils.idx2text(qi_i[0], options['vocabinv'])
for ii in range(prm.n_iterations):
prob = out.pop(0)
ans = out.pop(0)
metrics = out.pop(0)
bl = out.pop(0)
cost_bl = out.pop(0)
D_id = out.pop(0)
print
print 'Iteration', ii
print 'Baseline Value', bl.mean(), 'Cost', cost_bl
print ' '.join(prm.metrics_map.keys())
print metrics.mean(0)
print
print 'Retrieved Docs: ', str([options['engine'].id_title_map[d_id] for d_id in D_id[0]])
print
print 'Reformulated Query:', options['reformulated_queries'][ii][0]
print
print 'Query ANS: ',
for kk, word in enumerate(options['current_queries'][0][:ans.shape[1]]):
if word not in options['vocab'] and word != '':
word += '<unk>'
if ans[0,kk] == 1:
word = word.upper()
print str(word),
print
print
print 'prob[:,:,0].max(1).mean(), prob[:,:,0].mean(), prob[:,:,0].min(1).mean()', prob[:,:,0].max(1).mean(), prob[:,:,0].mean(), prob[:,:,0].min(1).mean()
print 'prob[:,:,1].max(1).mean(), prob[:,:,1].mean(), prob[:,:,1].min(1).mean()', prob[:,:,1].max(1).mean(), prob[:,:,1].mean(), prob[:,:,1].min(1).mean()
print '==================================================================================\n'
if np.mod(uidx, validFreq) == 0 or uidx == 1:
kf_train = get_minibatches_idx(len(qi_train), prm.batch_size_pred, shuffle=True, max_samples=train_size)
kf_valid = get_minibatches_idx(len(qi_valid), prm.batch_size_pred, shuffle=True, max_samples=valid_size)
kf_test = get_minibatches_idx(len(qi_test), prm.batch_size_pred, shuffle=True, max_samples=test_size)
print '\nEvaluating - Training Set'
train_metrics = pred_error(f_pred, qi_train, dt_train, options, kf_train)
print '\nEvaluating - Validation Set'
valid_metrics = pred_error(f_pred, qi_valid, dt_valid, options, kf_valid)
print '\nEvaluating - Test Set'
test_metrics = pred_error(f_pred, qi_test, dt_test, options, kf_test)
his = [train_metrics, valid_metrics, test_metrics]
history_errs.append(his)
metric_idx = prm.metrics_map[prm.reward.upper()]
if (uidx == 0 or
valid_metrics[-1, metric_idx] >= np.array(history_errs)[:,1,-1,metric_idx].max()):
best_p = unzip(tparams)
bad_counter = 0
print '====================================================================================================='
print ' '.join(prm.metrics_map.keys())
print
print 'Train:'
print train_metrics
print
print 'Valid:'
print valid_metrics
print
print 'Test:'
print test_metrics
print
print '====================================================================================================='
if (len(history_errs) > prm.patience and
valid_metrics[-1, metric_idx] <= np.array(history_errs)[:-prm.patience,
1,-1,metric_idx].max()):
bad_counter += 1
if bad_counter > prm.patience:
print 'Early Stop!'
estop = True
break
if prm.saveto and np.mod(uidx, saveFreq) == 0:
print 'Saving...',
if best_p is not None:
params = best_p
else:
params = unzip(tparams)
np.savez(prm.saveto, history_errs=history_errs, **params)
print 'Done'
print 'Seen %d samples' % n_samples
if estop:
break
except KeyboardInterrupt:
print "Training interupted"
return
if __name__ == '__main__':
# See parameters.py for all possible parameter and their definitions.
train()