chain-denominator.cc
19.9 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
// chain/chain-denominator.cc
// Copyright 2015 Johns Hopkins University (author: Daniel Povey)
// See ../../COPYING for clarification regarding multiple authors
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// THIS CODE IS PROVIDED *AS IS* BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
// KIND, EITHER EXPRESS OR IMPLIED, INCLUDING WITHOUT LIMITATION ANY IMPLIED
// WARRANTIES OR CONDITIONS OF TITLE, FITNESS FOR A PARTICULAR PURPOSE,
// MERCHANTABLITY OR NON-INFRINGEMENT.
// See the Apache 2 License for the specific language governing permissions and
// limitations under the License.
#include "chain/chain-denominator.h"
#include "chain/chain-kernels-ansi.h"
namespace kaldi {
namespace chain {
DenominatorComputation::DenominatorComputation(
const ChainTrainingOptions &opts,
const DenominatorGraph &den_graph,
int32 num_sequences,
const CuMatrixBase<BaseFloat> &nnet_output):
opts_(opts),
den_graph_(den_graph),
num_sequences_(num_sequences),
frames_per_sequence_(nnet_output.NumRows() / num_sequences_),
nnet_output_deriv_transposed_(
nnet_output.NumCols(),
std::min<int32>(nnet_output.NumRows(),
static_cast<int32>(kMaxDerivTimeSteps) *
num_sequences_)),
alpha_(frames_per_sequence_ + 1,
den_graph_.NumStates() * num_sequences_ + num_sequences_,
kUndefined),
beta_(2, den_graph_.NumStates() * num_sequences_ + num_sequences_,
kUndefined),
tot_prob_(num_sequences_, kUndefined),
tot_log_prob_(num_sequences_, kUndefined),
log_correction_term_(num_sequences_, kUndefined),
ok_(true) {
// We don't let leaky_hmm_coefficient be exactly zero (although that would
// make sense mathematically, corresponding to "turning off" the leaky HMM),
// because that would lead to underflow and eventually NaN's or inf's
// appearing in the computation, since we do this computation not in
// log-space.
KALDI_ASSERT(opts_.leaky_hmm_coefficient > 0.0 &&
opts_.leaky_hmm_coefficient < 1.0);
if (RandInt(0, 99) == 0) {
// A check, that all values in nnet_output are in the range [-30, 30]..
// otherwise derivatives will be wrong (search below for 30).
BaseFloat max_val = nnet_output.Max(), min_val = nnet_output.Min();
if (max_val > 30.0 || min_val < -30.0) {
KALDI_WARN << "Nnet outputs " << min_val << ", "
<< max_val <<
" outside the range [-30,30], derivs may be inaccurate.";
}
}
// make sure the alpha sums and beta sums are zeroed.
alpha_.ColRange(den_graph_.NumStates() * num_sequences_,
num_sequences_).SetZero();
beta_.ColRange(den_graph_.NumStates() * num_sequences_,
num_sequences_).SetZero();
KALDI_ASSERT(nnet_output.NumRows() % num_sequences == 0);
// the kStrideEqualNumCols argument is so that we can share the same
// memory block with xent_output_deriv (see chain-training.cc, search for
// kStrideEqualNumCols). This depends on how the allocator works, and
// actually might not happen, but anyway, the impact on speed would
// likely be un-measurably small.
exp_nnet_output_transposed_.Resize(nnet_output.NumCols(),
nnet_output.NumRows(),
kUndefined, kStrideEqualNumCols);
exp_nnet_output_transposed_.CopyFromMat(nnet_output, kTrans);
// We limit the nnet output to the range [-30,30] before doing the exp;
// this avoids NaNs appearing in the forward-backward computation, which
// is not done in log space.
exp_nnet_output_transposed_.ApplyExpLimited(-30.0, 30.0);
}
void DenominatorComputation::AlphaFirstFrame() {
// dim == num_hmm_states_ * num_sequences_.
BaseFloat *first_frame_alpha = alpha_.RowData(0);
// create a 'fake matrix' - view this row as a matrix.
// initializer takes [pointer, num-rows, num-cols, stride].
CuSubMatrix<BaseFloat> alpha_mat(first_frame_alpha,
den_graph_.NumStates(),
num_sequences_,
num_sequences_);
// TODO (possible): It would be more efficient here if we implemented a
// CopyColsFromVec function in class CuMatrix.
alpha_mat.SetZero();
alpha_mat.AddVecToCols(1.0, den_graph_.InitialProbs(), 0.0);
}
// the alpha computation for some 0 < t <= num_time_steps_.
void DenominatorComputation::AlphaGeneralFrame(int32 t) {
KALDI_ASSERT(t > 0 && t <= frames_per_sequence_);
BaseFloat *this_alpha = alpha_.RowData(t);
const BaseFloat *prev_alpha_dash = alpha_.RowData(t - 1);
const Int32Pair *backward_transitions = den_graph_.BackwardTransitions();
const DenominatorGraphTransition *transitions = den_graph_.Transitions();
int32 num_pdfs = exp_nnet_output_transposed_.NumRows(),
num_hmm_states = den_graph_.NumStates(),
num_sequences = num_sequences_;
// 'probs' is the matrix of pseudo-likelihoods for frame t - 1.
CuSubMatrix<BaseFloat> probs(exp_nnet_output_transposed_, 0, num_pdfs,
(t-1) * num_sequences_, num_sequences_);
const BaseFloat *prob_data = probs.Data();
#if HAVE_CUDA == 1
if (CuDevice::Instantiate().Enabled()) {
CuTimer tim;
dim3 dimBlock(std::min<int32>(CU1DBLOCK, num_sequences), 1, 1);
dim3 dimGrid(n_blocks(num_sequences, dimBlock.x), num_hmm_states, 1);
while (1) {
if (dimGrid.y > 65535) // the hardware doesn't allow more than this.
dimGrid.y = 65535;
cuda_chain_hmm_forward(dimGrid, dimBlock,
backward_transitions, transitions,
num_sequences, den_graph_.NumStates(),
prob_data, probs.Stride(), prev_alpha_dash,
this_alpha);
CU_SAFE_CALL(cudaGetLastError());
if (dimGrid.y == num_hmm_states) {
break; // this is the normal case.
} else {
// We reach this code only in the unusual case where num_hmm_states >
// 65535. We can compute the alphas for the remaining HMM states by
// moving some of the array pointers and making the call again.
backward_transitions += dimGrid.y;
this_alpha += dimGrid.y * num_sequences;
num_hmm_states -= dimGrid.y;
dimGrid.y = num_hmm_states;
}
}
CuDevice::Instantiate().AccuProfile(__func__, tim);
} else
#endif
{
int32 prob_stride = probs.Stride();
for (int32 h = 0; h < num_hmm_states; h++) {
for (int32 s = 0; s < num_sequences; s++) {
double this_tot_alpha = 0.0;
const DenominatorGraphTransition
*trans_iter = transitions + backward_transitions[h].first,
*trans_end = transitions + backward_transitions[h].second;
for (; trans_iter != trans_end; ++trans_iter) {
BaseFloat transition_prob = trans_iter->transition_prob;
int32 pdf_id = trans_iter->pdf_id,
prev_hmm_state = trans_iter->hmm_state;
BaseFloat prob = prob_data[pdf_id * prob_stride + s],
this_prev_alpha = prev_alpha_dash[prev_hmm_state * num_sequences + s];
this_tot_alpha += this_prev_alpha * transition_prob * prob;
}
// Let arbitrary_scale be the inverse of the alpha-sum value that we
// store in the same place we'd store the alpha for the state numbered
// 'num_hmm_states'. We multiply this into all the
// transition-probabilities from the previous frame to this frame, in
// both the forward and backward passes, in order to keep the alphas in
// a good numeric range. This won't affect the posteriors, but when
// computing the total likelihood we'll need to compensate for it later
// on.
BaseFloat arbitrary_scale =
1.0 / prev_alpha_dash[num_hmm_states * num_sequences + s];
KALDI_ASSERT(this_tot_alpha - this_tot_alpha == 0);
this_alpha[h * num_sequences + s] = this_tot_alpha * arbitrary_scale;
}
}
}
}
void DenominatorComputation::AlphaDash(int32 t) {
BaseFloat *this_alpha = alpha_.RowData(t);
// create a 'fake matrix' for the regular alphas- view this row as a matrix.
// initializer takes [pointer, num-rows, num-cols, stride].
CuSubMatrix<BaseFloat> alpha_mat(this_alpha,
den_graph_.NumStates(),
num_sequences_,
num_sequences_);
// the alpha-dash is the sum of alpha over all states.
CuSubVector<BaseFloat> alpha_sum_vec(this_alpha +
den_graph_.NumStates() * num_sequences_,
num_sequences_);
alpha_sum_vec.AddRowSumMat(1.0, alpha_mat, 0.0);
alpha_mat.AddVecVec(opts_.leaky_hmm_coefficient,
den_graph_.InitialProbs(),
alpha_sum_vec);
// it's now alpha-dash.
}
// compute beta from beta-dash.
void DenominatorComputation::Beta(int32 t) {
BaseFloat *this_beta_dash = beta_.RowData(t % 2);
// create a 'fake matrix' for the regular beta-dash (which is
// the counterpart of alpha-dash)- view this row as a matrix.
// initializer takes [pointer, num-rows, num-cols, stride].
CuSubMatrix<BaseFloat> beta_dash_mat(this_beta_dash,
den_graph_.NumStates(),
num_sequences_,
num_sequences_);
// making the t index implicit, the beta-dash-sum for each sequence is the sum
// over all states i of beta_i * opts_.leaky_hmm_coefficient * initial_prob_i.
CuSubVector<BaseFloat> beta_dash_sum_vec(
this_beta_dash + den_graph_.NumStates() * num_sequences_,
num_sequences_);
beta_dash_sum_vec.AddMatVec(opts_.leaky_hmm_coefficient, beta_dash_mat,
kTrans, den_graph_.InitialProbs(), 0.0);
// we are computing beta in place. After the following, beta-dash-mat
// will contain the actual beta (i.e. the counterpart of alpha),
// not the beta-dash.
beta_dash_mat.AddVecToRows(1.0, beta_dash_sum_vec);
}
BaseFloat DenominatorComputation::Forward() {
AlphaFirstFrame();
AlphaDash(0);
for (int32 t = 1; t <= frames_per_sequence_; t++) {
AlphaGeneralFrame(t);
AlphaDash(t);
}
return ComputeTotLogLike();
}
BaseFloat DenominatorComputation::ComputeTotLogLike() {
tot_prob_.Resize(num_sequences_);
// View the last alpha-dash as a matrix of size num-hmm-states by num-sequences.
CuSubMatrix<BaseFloat> last_alpha_dash(
alpha_.RowData(frames_per_sequence_),
den_graph_.NumStates(),
num_sequences_,
num_sequences_);
tot_prob_.AddRowSumMat(1.0, last_alpha_dash, 0.0);
// we should probably add an ApplyLog() function that takes a vector argument.
tot_log_prob_ = tot_prob_;
tot_log_prob_.ApplyLog();
BaseFloat tot_log_prob = tot_log_prob_.Sum();
// We now have to add something for the arbitrary scaling factor. [note: the
// purpose of the arbitrary scaling factors was to keep things in a good
// floating-point range]
// The inverses of all the tot-alpha quantities, for t = 0
// ... frames_per_sequence_ - 1, were included as the 'arbitrary factors' in
// the transition-probs, so we need to multiply them all together (not
// inversed) and add them as a correction term to the total log-likes.
// These tot-alpha quantities were stored in the same place that we would
// have stored the HMM-state numbered 'num_hmm_states'.
int32 num_hmm_states = den_graph_.NumStates();
CuSubMatrix<BaseFloat> inv_arbitrary_scales(
alpha_, 0, frames_per_sequence_,
num_sequences_ * num_hmm_states, num_sequences_);
CuMatrix<BaseFloat> log_inv_arbitrary_scales(
inv_arbitrary_scales);
log_inv_arbitrary_scales.ApplyLog();
BaseFloat log_inv_arbitrary_scales_product =
log_inv_arbitrary_scales.Sum();
return tot_log_prob + log_inv_arbitrary_scales_product;
}
bool DenominatorComputation::Backward(
BaseFloat deriv_weight,
CuMatrixBase<BaseFloat> *nnet_output_deriv) {
BetaDashLastFrame();
Beta(frames_per_sequence_);
for (int32 t = frames_per_sequence_ - 1; t >= 0; t--) {
BetaDashGeneralFrame(t);
if (GetVerboseLevel() >= 1 || t == 0)
BetaGeneralFrameDebug(t);
Beta(t);
if (t % kMaxDerivTimeSteps == 0) {
// commit the derivative stored in nnet_output_deriv_transposed_ by adding
// its transpose to the appropriate sub-matrix of 'nnet_output_deriv'.
int32 chunk_frames = std::min<int32>(static_cast<int32>(kMaxDerivTimeSteps),
frames_per_sequence_ - t),
num_pdfs = exp_nnet_output_transposed_.NumRows();
CuSubMatrix<BaseFloat> transposed_deriv_part(
nnet_output_deriv_transposed_,
0, num_pdfs,
0, chunk_frames * num_sequences_);
CuSubMatrix<BaseFloat> output_deriv_part(
*nnet_output_deriv,
t * num_sequences_, chunk_frames * num_sequences_,
0, num_pdfs);
output_deriv_part.AddMat(deriv_weight, transposed_deriv_part, kTrans);
if (t != 0)
transposed_deriv_part.SetZero();
}
}
return ok_;
}
void DenominatorComputation::BetaDashLastFrame() {
// sets up the beta-dash quantity on the last frame (frame ==
// frames_per_sequence_). Note that the betas we use here contain a
// 1/(tot-prob) factor in order to simplify the backprop.
int32 t = frames_per_sequence_;
BaseFloat *last_frame_beta_dash = beta_.RowData(t % 2);
// create a 'fake matrix' - view this row as a matrix.
CuSubMatrix<BaseFloat> beta_dash_mat(last_frame_beta_dash,
den_graph_.NumStates(),
num_sequences_,
num_sequences_);
CuVector<BaseFloat> inv_tot_prob(tot_prob_);
inv_tot_prob.InvertElements();
// the beta values at the end of the file only vary with the sequence-index,
// not with the HMM-index. We treat all states as having a final-prob of one.
beta_dash_mat.CopyRowsFromVec(inv_tot_prob);
}
void DenominatorComputation::BetaDashGeneralFrame(int32 t) {
KALDI_ASSERT(t >= 0 && t < frames_per_sequence_);
int32 num_pdfs = exp_nnet_output_transposed_.NumRows();
// t_wrapped gives us the time-index we use when indexing
// nnet_output_deriv_transposed_; to save memory we limit the size of the
// matrix, storing only chunks of frames at a time, and we add it to the
// non-transposed output whenever we finish a chunk.
int32 t_wrapped = t % static_cast<int32>(kMaxDerivTimeSteps);
const BaseFloat *this_alpha_dash = alpha_.RowData(t),
*next_beta = beta_.RowData((t + 1) % 2);
BaseFloat *this_beta_dash = beta_.RowData(t % 2);
const Int32Pair *forward_transitions = den_graph_.ForwardTransitions();
const DenominatorGraphTransition *transitions = den_graph_.Transitions();
// 'probs' is the matrix of pseudo-likelihoods for frame t.
CuSubMatrix<BaseFloat> probs(exp_nnet_output_transposed_, 0, num_pdfs,
t * num_sequences_, num_sequences_),
log_prob_deriv(nnet_output_deriv_transposed_, 0, num_pdfs,
t_wrapped * num_sequences_, num_sequences_);
int32 num_hmm_states = den_graph_.NumStates(),
num_sequences = num_sequences_;
#if HAVE_CUDA == 1
if (CuDevice::Instantiate().Enabled()) {
CuTimer tim;
dim3 dimBlock(std::min<int32>(CU1DBLOCK, num_sequences), 1, 1);
dim3 dimGrid(n_blocks(num_sequences, dimBlock.x), num_hmm_states, 1);
while (1) {
if (dimGrid.y > 65535) // the hardware doesn't allow more than this.
dimGrid.y = 65535;
cuda_chain_hmm_backward(dimGrid, dimBlock, forward_transitions, transitions,
num_sequences, num_hmm_states,
probs.Data(), probs.Stride(),
this_alpha_dash, next_beta, this_beta_dash,
log_prob_deriv.Data(), log_prob_deriv.Stride());
CU_SAFE_CALL(cudaGetLastError());
if (dimGrid.y == num_hmm_states) {
break; // this is the normal case.
} else {
// We reach this code only in the unusual case where num_hmm_states >
// 65535. We can compute the betas (and log-prob derivatives) for the
// remaining HMM states by moving some of the array pointers and making
// the call again.
forward_transitions += dimGrid.y;
this_alpha_dash += dimGrid.y * num_sequences;
this_beta_dash += dimGrid.y * num_sequences;
num_hmm_states -= dimGrid.y;
dimGrid.y = num_hmm_states;
}
}
CuDevice::Instantiate().AccuProfile(__func__, tim);
} else
#endif
{
int32 prob_stride = probs.Stride(),
deriv_stride = log_prob_deriv.Stride();
const BaseFloat *prob_data = probs.Data();
BaseFloat *log_prob_deriv_data = log_prob_deriv.Data();
for (int32 h = 0; h < num_hmm_states; h++) {
for (int32 s = 0; s < num_sequences; s++) {
BaseFloat this_alpha_dash_prob = this_alpha_dash[h * num_sequences + s],
inv_arbitrary_scale =
this_alpha_dash[num_hmm_states * num_sequences + s];
double tot_variable_factor = 0.0;
BaseFloat occupation_factor = this_alpha_dash_prob /
inv_arbitrary_scale;
const DenominatorGraphTransition
*trans_iter = transitions + forward_transitions[h].first,
*trans_end = transitions + forward_transitions[h].second;
for (; trans_iter != trans_end; ++trans_iter) {
BaseFloat transition_prob = trans_iter->transition_prob;
int32 pdf_id = trans_iter->pdf_id,
next_hmm_state = trans_iter->hmm_state;
BaseFloat variable_factor = transition_prob *
next_beta[next_hmm_state * num_sequences + s] *
prob_data[pdf_id * prob_stride + s];
tot_variable_factor += variable_factor;
BaseFloat occupation_prob = variable_factor * occupation_factor;
log_prob_deriv_data[pdf_id * deriv_stride + s] += occupation_prob;
}
this_beta_dash[h * num_sequences + s] =
tot_variable_factor / inv_arbitrary_scale;
}
}
}
}
void DenominatorComputation::BetaGeneralFrameDebug(int32 t) {
BaseFloat num_hmm_states = den_graph_.NumStates(),
alpha_beta_size = num_hmm_states * num_sequences_;
CuSubVector<BaseFloat> this_alpha_dash(alpha_.RowData(t), alpha_beta_size),
this_beta_dash(beta_.RowData(t % 2), alpha_beta_size);
int32 t_wrapped = t % static_cast<int32>(kMaxDerivTimeSteps),
num_pdfs = exp_nnet_output_transposed_.NumRows();
CuSubMatrix<BaseFloat> this_log_prob_deriv(
nnet_output_deriv_transposed_, 0, num_pdfs,
t_wrapped * num_sequences_, num_sequences_);
BaseFloat alpha_beta_product = VecVec(this_alpha_dash,
this_beta_dash),
this_log_prob_deriv_sum = this_log_prob_deriv.Sum();
if (!ApproxEqual(alpha_beta_product, num_sequences_)) {
KALDI_WARN << "On time " << t << ", alpha-beta product "
<< alpha_beta_product << " != " << num_sequences_
<< " alpha-dash-sum = " << this_alpha_dash.Sum()
<< ", beta-dash-sum = " << this_beta_dash.Sum();
if (fabs(alpha_beta_product - num_sequences_) > 2.0) {
KALDI_WARN << "Excessive error detected, will abandon this minibatch";
ok_ = false;
}
}
// use higher tolerance, since we are using randomized pruning for the
// log-prob derivatives.
if (!ApproxEqual(this_log_prob_deriv_sum,
num_sequences_, 0.01)) {
KALDI_WARN << "On time " << t << ", log-prob-deriv sum "
<< this_log_prob_deriv_sum << " != " << num_sequences_;
if (fabs(this_log_prob_deriv_sum - num_sequences_) > 2.0) {
KALDI_WARN << "Excessive error detected, will abandon this minibatch";
ok_ = false;
}
}
}
} // namespace chain
} // namespace kaldi