/
regression_obj.cu
584 lines (513 loc) · 21.2 KB
/
regression_obj.cu
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
/*!
* Copyright 2015-2019 by Contributors
* \file regression_obj.cu
* \brief Definition of single-value regression and classification objectives.
* \author Tianqi Chen, Kailong Chen
*/
#include <dmlc/omp.h>
#include <xgboost/logging.h>
#include <xgboost/objective.h>
#include <cmath>
#include <memory>
#include <vector>
#include "xgboost/host_device_vector.h"
#include "xgboost/json.h"
#include "xgboost/parameter.h"
#include "xgboost/span.h"
#include "../common/transform.h"
#include "../common/common.h"
#include "../common/threading_utils.h"
#include "./regression_loss.h"
namespace xgboost {
namespace obj {
#if defined(XGBOOST_USE_CUDA)
DMLC_REGISTRY_FILE_TAG(regression_obj_gpu);
#endif // defined(XGBOOST_USE_CUDA)
struct RegLossParam : public XGBoostParameter<RegLossParam> {
float scale_pos_weight;
// declare parameters
DMLC_DECLARE_PARAMETER(RegLossParam) {
DMLC_DECLARE_FIELD(scale_pos_weight).set_default(1.0f).set_lower_bound(0.0f)
.describe("Scale the weight of positive examples by this factor");
}
};
template<typename Loss>
class RegLossObj : public ObjFunction {
protected:
HostDeviceVector<float> additional_input_;
public:
// 0 - label_correct flag, 1 - scale_pos_weight, 2 - is_null_weight
RegLossObj(): additional_input_(3) {}
void Configure(const std::vector<std::pair<std::string, std::string> >& args) override {
param_.UpdateAllowUnknown(args);
}
void GetGradient(const HostDeviceVector<bst_float>& preds,
const MetaInfo &info, int,
HostDeviceVector<GradientPair>* out_gpair) override {
CHECK_EQ(preds.Size(), info.labels_.Size())
<< " " << "labels are not correctly provided"
<< "preds.size=" << preds.Size() << ", label.size=" << info.labels_.Size() << ", "
<< "Loss: " << Loss::Name();
size_t const ndata = preds.Size();
out_gpair->Resize(ndata);
auto device = tparam_->gpu_id;
additional_input_.HostVector().begin()[0] = 1; // Fill the label_correct flag
bool is_null_weight = info.weights_.Size() == 0;
if (!is_null_weight) {
CHECK_EQ(info.weights_.Size(), ndata)
<< "Number of weights should be equal to number of data points.";
}
auto scale_pos_weight = param_.scale_pos_weight;
additional_input_.HostVector().begin()[1] = scale_pos_weight;
additional_input_.HostVector().begin()[2] = is_null_weight;
const size_t nthreads = tparam_->Threads();
bool on_device = device >= 0;
// On CPU we run the transformation each thread processing a contigious block of data
// for better performance.
const size_t n_data_blocks =
std::max(static_cast<size_t>(1), (on_device ? ndata : nthreads));
const size_t block_size = ndata / n_data_blocks + !!(ndata % n_data_blocks);
common::Transform<>::Init(
[block_size, ndata] XGBOOST_DEVICE(
size_t data_block_idx, common::Span<float> _additional_input,
common::Span<GradientPair> _out_gpair,
common::Span<const bst_float> _preds,
common::Span<const bst_float> _labels,
common::Span<const bst_float> _weights) {
const bst_float* preds_ptr = _preds.data();
const bst_float* labels_ptr = _labels.data();
const bst_float* weights_ptr = _weights.data();
GradientPair* out_gpair_ptr = _out_gpair.data();
const size_t begin = data_block_idx*block_size;
const size_t end = std::min(ndata, begin + block_size);
const float _scale_pos_weight = _additional_input[1];
const bool _is_null_weight = _additional_input[2];
for (size_t idx = begin; idx < end; ++idx) {
bst_float p = Loss::PredTransform(preds_ptr[idx]);
bst_float w = _is_null_weight ? 1.0f : weights_ptr[idx];
bst_float label = labels_ptr[idx];
if (label == 1.0f) {
w *= _scale_pos_weight;
}
if (!Loss::CheckLabel(label)) {
// If there is an incorrect label, the host code will know.
_additional_input[0] = 0;
}
out_gpair_ptr[idx] = GradientPair(Loss::FirstOrderGradient(p, label) * w,
Loss::SecondOrderGradient(p, label) * w);
}
},
common::Range{0, static_cast<int64_t>(n_data_blocks)}, device)
.Eval(&additional_input_, out_gpair, &preds, &info.labels_,
&info.weights_);
auto const flag = additional_input_.HostVector().begin()[0];
if (flag == 0) {
LOG(FATAL) << Loss::LabelErrorMsg();
}
}
public:
const char* DefaultEvalMetric() const override {
return Loss::DefaultEvalMetric();
}
void PredTransform(HostDeviceVector<float> *io_preds) const override {
common::Transform<>::Init(
[] XGBOOST_DEVICE(size_t _idx, common::Span<float> _preds) {
_preds[_idx] = Loss::PredTransform(_preds[_idx]);
}, common::Range{0, static_cast<int64_t>(io_preds->Size())},
io_preds->DeviceIdx())
.Eval(io_preds);
}
float ProbToMargin(float base_score) const override {
return Loss::ProbToMargin(base_score);
}
void SaveConfig(Json* p_out) const override {
auto& out = *p_out;
out["name"] = String(Loss::Name());
out["reg_loss_param"] = ToJson(param_);
}
void LoadConfig(Json const& in) override {
FromJson(in["reg_loss_param"], ¶m_);
}
protected:
RegLossParam param_;
};
// register the objective functions
DMLC_REGISTER_PARAMETER(RegLossParam);
XGBOOST_REGISTER_OBJECTIVE(SquaredLossRegression, LinearSquareLoss::Name())
.describe("Regression with squared error.")
.set_body([]() { return new RegLossObj<LinearSquareLoss>(); });
XGBOOST_REGISTER_OBJECTIVE(SquareLogError, SquaredLogError::Name())
.describe("Regression with root mean squared logarithmic error.")
.set_body([]() { return new RegLossObj<SquaredLogError>(); });
XGBOOST_REGISTER_OBJECTIVE(LogisticRegression, LogisticRegression::Name())
.describe("Logistic regression for probability regression task.")
.set_body([]() { return new RegLossObj<LogisticRegression>(); });
XGBOOST_REGISTER_OBJECTIVE(PseudoHuberError, PseudoHuberError::Name())
.describe("Regression Pseudo Huber error.")
.set_body([]() { return new RegLossObj<PseudoHuberError>(); });
XGBOOST_REGISTER_OBJECTIVE(LogisticClassification, LogisticClassification::Name())
.describe("Logistic regression for binary classification task.")
.set_body([]() { return new RegLossObj<LogisticClassification>(); });
XGBOOST_REGISTER_OBJECTIVE(LogisticRaw, LogisticRaw::Name())
.describe("Logistic regression for classification, output score "
"before logistic transformation.")
.set_body([]() { return new RegLossObj<LogisticRaw>(); });
// Deprecated functions
XGBOOST_REGISTER_OBJECTIVE(LinearRegression, "reg:linear")
.describe("Regression with squared error.")
.set_body([]() {
LOG(WARNING) << "reg:linear is now deprecated in favor of reg:squarederror.";
return new RegLossObj<LinearSquareLoss>(); });
// End deprecated
// declare parameter
struct PoissonRegressionParam : public XGBoostParameter<PoissonRegressionParam> {
float max_delta_step;
DMLC_DECLARE_PARAMETER(PoissonRegressionParam) {
DMLC_DECLARE_FIELD(max_delta_step).set_lower_bound(0.0f).set_default(0.7f)
.describe("Maximum delta step we allow each weight estimation to be." \
" This parameter is required for possion regression.");
}
};
// poisson regression for count
class PoissonRegression : public ObjFunction {
public:
// declare functions
void Configure(const std::vector<std::pair<std::string, std::string> >& args) override {
param_.UpdateAllowUnknown(args);
}
void GetGradient(const HostDeviceVector<bst_float>& preds,
const MetaInfo &info, int,
HostDeviceVector<GradientPair> *out_gpair) override {
CHECK_NE(info.labels_.Size(), 0U) << "label set cannot be empty";
CHECK_EQ(preds.Size(), info.labels_.Size()) << "labels are not correctly provided";
size_t const ndata = preds.Size();
out_gpair->Resize(ndata);
auto device = tparam_->gpu_id;
label_correct_.Resize(1);
label_correct_.Fill(1);
bool is_null_weight = info.weights_.Size() == 0;
if (!is_null_weight) {
CHECK_EQ(info.weights_.Size(), ndata)
<< "Number of weights should be equal to number of data points.";
}
bst_float max_delta_step = param_.max_delta_step;
common::Transform<>::Init(
[=] XGBOOST_DEVICE(size_t _idx,
common::Span<int> _label_correct,
common::Span<GradientPair> _out_gpair,
common::Span<const bst_float> _preds,
common::Span<const bst_float> _labels,
common::Span<const bst_float> _weights) {
bst_float p = _preds[_idx];
bst_float w = is_null_weight ? 1.0f : _weights[_idx];
bst_float y = _labels[_idx];
if (y < 0.0f) {
_label_correct[0] = 0;
}
_out_gpair[_idx] = GradientPair{(expf(p) - y) * w,
expf(p + max_delta_step) * w};
},
common::Range{0, static_cast<int64_t>(ndata)}, device).Eval(
&label_correct_, out_gpair, &preds, &info.labels_, &info.weights_);
// copy "label correct" flags back to host
std::vector<int>& label_correct_h = label_correct_.HostVector();
for (auto const flag : label_correct_h) {
if (flag == 0) {
LOG(FATAL) << "PoissonRegression: label must be nonnegative";
}
}
}
void PredTransform(HostDeviceVector<bst_float> *io_preds) const override {
common::Transform<>::Init(
[] XGBOOST_DEVICE(size_t _idx, common::Span<bst_float> _preds) {
_preds[_idx] = expf(_preds[_idx]);
},
common::Range{0, static_cast<int64_t>(io_preds->Size())},
io_preds->DeviceIdx())
.Eval(io_preds);
}
void EvalTransform(HostDeviceVector<bst_float> *io_preds) override {
PredTransform(io_preds);
}
bst_float ProbToMargin(bst_float base_score) const override {
return std::log(base_score);
}
const char* DefaultEvalMetric() const override {
return "poisson-nloglik";
}
void SaveConfig(Json* p_out) const override {
auto& out = *p_out;
out["name"] = String("count:poisson");
out["poisson_regression_param"] = ToJson(param_);
}
void LoadConfig(Json const& in) override {
FromJson(in["poisson_regression_param"], ¶m_);
}
private:
PoissonRegressionParam param_;
HostDeviceVector<int> label_correct_;
};
// register the objective functions
DMLC_REGISTER_PARAMETER(PoissonRegressionParam);
XGBOOST_REGISTER_OBJECTIVE(PoissonRegression, "count:poisson")
.describe("Poisson regression for count data.")
.set_body([]() { return new PoissonRegression(); });
// cox regression for survival data (negative values mean they are censored)
class CoxRegression : public ObjFunction {
public:
void Configure(
const std::vector<std::pair<std::string, std::string> >&) override {}
void GetGradient(const HostDeviceVector<bst_float>& preds,
const MetaInfo &info, int,
HostDeviceVector<GradientPair> *out_gpair) override {
CHECK_NE(info.labels_.Size(), 0U) << "label set cannot be empty";
CHECK_EQ(preds.Size(), info.labels_.Size()) << "labels are not correctly provided";
const auto& preds_h = preds.HostVector();
out_gpair->Resize(preds_h.size());
auto& gpair = out_gpair->HostVector();
const std::vector<size_t> &label_order = info.LabelAbsSort();
const omp_ulong ndata = static_cast<omp_ulong>(preds_h.size()); // NOLINT(*)
const bool is_null_weight = info.weights_.Size() == 0;
if (!is_null_weight) {
CHECK_EQ(info.weights_.Size(), ndata)
<< "Number of weights should be equal to number of data points.";
}
// pre-compute a sum
double exp_p_sum = 0; // we use double because we might need the precision with large datasets
for (omp_ulong i = 0; i < ndata; ++i) {
exp_p_sum += std::exp(preds_h[label_order[i]]);
}
// start calculating grad and hess
const auto& labels = info.labels_.HostVector();
double r_k = 0;
double s_k = 0;
double last_exp_p = 0.0;
double last_abs_y = 0.0;
double accumulated_sum = 0;
for (omp_ulong i = 0; i < ndata; ++i) { // NOLINT(*)
const size_t ind = label_order[i];
const double p = preds_h[ind];
const double exp_p = std::exp(p);
const double w = info.GetWeight(ind);
const double y = labels[ind];
const double abs_y = std::abs(y);
// only update the denominator after we move forward in time (labels are sorted)
// this is Breslow's method for ties
accumulated_sum += last_exp_p;
if (last_abs_y < abs_y) {
exp_p_sum -= accumulated_sum;
accumulated_sum = 0;
} else {
CHECK(last_abs_y <= abs_y) << "CoxRegression: labels must be in sorted order, " <<
"MetaInfo::LabelArgsort failed!";
}
if (y > 0) {
r_k += 1.0/exp_p_sum;
s_k += 1.0/(exp_p_sum*exp_p_sum);
}
const double grad = exp_p*r_k - static_cast<bst_float>(y > 0);
const double hess = exp_p*r_k - exp_p*exp_p * s_k;
gpair.at(ind) = GradientPair(grad * w, hess * w);
last_abs_y = abs_y;
last_exp_p = exp_p;
}
}
void PredTransform(HostDeviceVector<bst_float> *io_preds) const override {
std::vector<bst_float> &preds = io_preds->HostVector();
const long ndata = static_cast<long>(preds.size()); // NOLINT(*)
common::ParallelFor(ndata, [&](long j) { // NOLINT(*)
preds[j] = std::exp(preds[j]);
});
}
void EvalTransform(HostDeviceVector<bst_float> *io_preds) override {
PredTransform(io_preds);
}
bst_float ProbToMargin(bst_float base_score) const override {
return std::log(base_score);
}
const char* DefaultEvalMetric() const override {
return "cox-nloglik";
}
void SaveConfig(Json* p_out) const override {
auto& out = *p_out;
out["name"] = String("survival:cox");
}
void LoadConfig(Json const&) override {}
};
// register the objective function
XGBOOST_REGISTER_OBJECTIVE(CoxRegression, "survival:cox")
.describe("Cox regression for censored survival data (negative labels are considered censored).")
.set_body([]() { return new CoxRegression(); });
// gamma regression
class GammaRegression : public ObjFunction {
public:
void Configure(
const std::vector<std::pair<std::string, std::string> >&) override {}
void GetGradient(const HostDeviceVector<bst_float> &preds,
const MetaInfo &info, int,
HostDeviceVector<GradientPair> *out_gpair) override {
CHECK_NE(info.labels_.Size(), 0U) << "label set cannot be empty";
CHECK_EQ(preds.Size(), info.labels_.Size()) << "labels are not correctly provided";
const size_t ndata = preds.Size();
auto device = tparam_->gpu_id;
out_gpair->Resize(ndata);
label_correct_.Resize(1);
label_correct_.Fill(1);
const bool is_null_weight = info.weights_.Size() == 0;
if (!is_null_weight) {
CHECK_EQ(info.weights_.Size(), ndata)
<< "Number of weights should be equal to number of data points.";
}
common::Transform<>::Init(
[=] XGBOOST_DEVICE(size_t _idx,
common::Span<int> _label_correct,
common::Span<GradientPair> _out_gpair,
common::Span<const bst_float> _preds,
common::Span<const bst_float> _labels,
common::Span<const bst_float> _weights) {
bst_float p = _preds[_idx];
bst_float w = is_null_weight ? 1.0f : _weights[_idx];
bst_float y = _labels[_idx];
if (y < 0.0f) {
_label_correct[0] = 0;
}
_out_gpair[_idx] = GradientPair((1 - y / expf(p)) * w, y / expf(p) * w);
},
common::Range{0, static_cast<int64_t>(ndata)}, device).Eval(
&label_correct_, out_gpair, &preds, &info.labels_, &info.weights_);
// copy "label correct" flags back to host
std::vector<int>& label_correct_h = label_correct_.HostVector();
for (auto const flag : label_correct_h) {
if (flag == 0) {
LOG(FATAL) << "GammaRegression: label must be non-negative.";
}
}
}
void PredTransform(HostDeviceVector<bst_float> *io_preds) const override {
common::Transform<>::Init(
[] XGBOOST_DEVICE(size_t _idx, common::Span<bst_float> _preds) {
_preds[_idx] = expf(_preds[_idx]);
},
common::Range{0, static_cast<int64_t>(io_preds->Size())},
io_preds->DeviceIdx())
.Eval(io_preds);
}
void EvalTransform(HostDeviceVector<bst_float> *io_preds) override {
PredTransform(io_preds);
}
bst_float ProbToMargin(bst_float base_score) const override {
return std::log(base_score);
}
const char* DefaultEvalMetric() const override {
return "gamma-nloglik";
}
void SaveConfig(Json* p_out) const override {
auto& out = *p_out;
out["name"] = String("reg:gamma");
}
void LoadConfig(Json const&) override {}
private:
HostDeviceVector<int> label_correct_;
};
// register the objective functions
XGBOOST_REGISTER_OBJECTIVE(GammaRegression, "reg:gamma")
.describe("Gamma regression for severity data.")
.set_body([]() { return new GammaRegression(); });
// declare parameter
struct TweedieRegressionParam : public XGBoostParameter<TweedieRegressionParam> {
float tweedie_variance_power;
DMLC_DECLARE_PARAMETER(TweedieRegressionParam) {
DMLC_DECLARE_FIELD(tweedie_variance_power).set_range(1.0f, 2.0f).set_default(1.5f)
.describe("Tweedie variance power. Must be between in range [1, 2).");
}
};
// tweedie regression
class TweedieRegression : public ObjFunction {
public:
// declare functions
void Configure(const std::vector<std::pair<std::string, std::string> >& args) override {
param_.UpdateAllowUnknown(args);
std::ostringstream os;
os << "tweedie-nloglik@" << param_.tweedie_variance_power;
metric_ = os.str();
}
void GetGradient(const HostDeviceVector<bst_float>& preds,
const MetaInfo &info, int,
HostDeviceVector<GradientPair> *out_gpair) override {
CHECK_NE(info.labels_.Size(), 0U) << "label set cannot be empty";
CHECK_EQ(preds.Size(), info.labels_.Size()) << "labels are not correctly provided";
const size_t ndata = preds.Size();
out_gpair->Resize(ndata);
auto device = tparam_->gpu_id;
label_correct_.Resize(1);
label_correct_.Fill(1);
const bool is_null_weight = info.weights_.Size() == 0;
if (!is_null_weight) {
CHECK_EQ(info.weights_.Size(), ndata)
<< "Number of weights should be equal to number of data points.";
}
const float rho = param_.tweedie_variance_power;
common::Transform<>::Init(
[=] XGBOOST_DEVICE(size_t _idx,
common::Span<int> _label_correct,
common::Span<GradientPair> _out_gpair,
common::Span<const bst_float> _preds,
common::Span<const bst_float> _labels,
common::Span<const bst_float> _weights) {
bst_float p = _preds[_idx];
bst_float w = is_null_weight ? 1.0f : _weights[_idx];
bst_float y = _labels[_idx];
if (y < 0.0f) {
_label_correct[0] = 0;
}
bst_float grad = -y * expf((1 - rho) * p) + expf((2 - rho) * p);
bst_float hess =
-y * (1 - rho) * \
std::exp((1 - rho) * p) + (2 - rho) * expf((2 - rho) * p);
_out_gpair[_idx] = GradientPair(grad * w, hess * w);
},
common::Range{0, static_cast<int64_t>(ndata), 1}, device)
.Eval(&label_correct_, out_gpair, &preds, &info.labels_, &info.weights_);
// copy "label correct" flags back to host
std::vector<int>& label_correct_h = label_correct_.HostVector();
for (auto const flag : label_correct_h) {
if (flag == 0) {
LOG(FATAL) << "TweedieRegression: label must be nonnegative";
}
}
}
void PredTransform(HostDeviceVector<bst_float> *io_preds) const override {
common::Transform<>::Init(
[] XGBOOST_DEVICE(size_t _idx, common::Span<bst_float> _preds) {
_preds[_idx] = expf(_preds[_idx]);
},
common::Range{0, static_cast<int64_t>(io_preds->Size())},
io_preds->DeviceIdx())
.Eval(io_preds);
}
bst_float ProbToMargin(bst_float base_score) const override {
return std::log(base_score);
}
const char* DefaultEvalMetric() const override {
return metric_.c_str();
}
void SaveConfig(Json* p_out) const override {
auto& out = *p_out;
out["name"] = String("reg:tweedie");
out["tweedie_regression_param"] = ToJson(param_);
}
void LoadConfig(Json const& in) override {
FromJson(in["tweedie_regression_param"], ¶m_);
}
private:
std::string metric_;
TweedieRegressionParam param_;
HostDeviceVector<int> label_correct_;
};
// register the objective functions
DMLC_REGISTER_PARAMETER(TweedieRegressionParam);
XGBOOST_REGISTER_OBJECTIVE(TweedieRegression, "reg:tweedie")
.describe("Tweedie regression for insurance data.")
.set_body([]() { return new TweedieRegression(); });
} // namespace obj
} // namespace xgboost