tree.cpp 41.9 KB
Newer Older
1
2
3
4
/*!
 * Copyright (c) 2016 Microsoft Corporation. All rights reserved.
 * Licensed under the MIT License. See LICENSE file in the project root for license information.
 */
Guolin Ke's avatar
Guolin Ke committed
5
6
7
#include <LightGBM/tree.h>

#include <LightGBM/dataset.h>
8
9
#include <LightGBM/utils/common.h>
#include <LightGBM/utils/threading.h>
Guolin Ke's avatar
Guolin Ke committed
10

11
12
13
14
#include <functional>
#include <iomanip>
#include <sstream>

Guolin Ke's avatar
Guolin Ke committed
15
16
namespace LightGBM {

17
Tree::Tree(int max_leaves, bool track_branch_features, bool is_linear)
18
  :max_leaves_(max_leaves), track_branch_features_(track_branch_features) {
Guolin Ke's avatar
Guolin Ke committed
19
20
21
22
23
24
  left_child_.resize(max_leaves_ - 1);
  right_child_.resize(max_leaves_ - 1);
  split_feature_inner_.resize(max_leaves_ - 1);
  split_feature_.resize(max_leaves_ - 1);
  threshold_in_bin_.resize(max_leaves_ - 1);
  threshold_.resize(max_leaves_ - 1);
Guolin Ke's avatar
Guolin Ke committed
25
  decision_type_.resize(max_leaves_ - 1, 0);
Guolin Ke's avatar
Guolin Ke committed
26
27
28
  split_gain_.resize(max_leaves_ - 1);
  leaf_parent_.resize(max_leaves_);
  leaf_value_.resize(max_leaves_);
29
  leaf_weight_.resize(max_leaves_);
Guolin Ke's avatar
Guolin Ke committed
30
31
  leaf_count_.resize(max_leaves_);
  internal_value_.resize(max_leaves_ - 1);
32
  internal_weight_.resize(max_leaves_ - 1);
Guolin Ke's avatar
Guolin Ke committed
33
34
  internal_count_.resize(max_leaves_ - 1);
  leaf_depth_.resize(max_leaves_);
35
36
37
  if (track_branch_features_) {
    branch_features_ = std::vector<std::vector<int>>(max_leaves_);
  }
Guolin Ke's avatar
Guolin Ke committed
38
39
  // root is in the depth 0
  leaf_depth_[0] = 0;
Guolin Ke's avatar
Guolin Ke committed
40
  num_leaves_ = 1;
41
  leaf_value_[0] = 0.0f;
42
  leaf_weight_[0] = 0.0f;
Guolin Ke's avatar
Guolin Ke committed
43
  leaf_parent_[0] = -1;
Guolin Ke's avatar
Guolin Ke committed
44
  shrinkage_ = 1.0f;
45
  num_cat_ = 0;
46
47
  cat_boundaries_.push_back(0);
  cat_boundaries_inner_.push_back(0);
48
  max_depth_ = -1;
49
50
51
52
53
54
55
  is_linear_ = is_linear;
  if (is_linear_) {
    leaf_coeff_.resize(max_leaves_);
    leaf_const_ = std::vector<double>(max_leaves_, 0);
    leaf_features_.resize(max_leaves_);
    leaf_features_inner_.resize(max_leaves_);
  }
Guolin Ke's avatar
Guolin Ke committed
56
}
Guolin Ke's avatar
Guolin Ke committed
57

58
59
int Tree::Split(int leaf, int feature, int real_feature, uint32_t threshold_bin,
                double threshold_double, double left_value, double right_value,
60
61
                int left_cnt, int right_cnt, double left_weight, double right_weight, float gain,
                MissingType missing_type, bool default_left) {
62
  Split(leaf, feature, real_feature, left_value, right_value, left_cnt, right_cnt, left_weight, right_weight, gain);
Guolin Ke's avatar
Guolin Ke committed
63
  int new_node_idx = num_leaves_ - 1;
Guolin Ke's avatar
Guolin Ke committed
64
  decision_type_[new_node_idx] = 0;
65
  SetDecisionType(&decision_type_[new_node_idx], false, kCategoricalMask);
Guolin Ke's avatar
Guolin Ke committed
66
  SetDecisionType(&decision_type_[new_node_idx], default_left, kDefaultLeftMask);
Guolin Ke's avatar
Guolin Ke committed
67
  SetMissingType(&decision_type_[new_node_idx], static_cast<int8_t>(missing_type));
Guolin Ke's avatar
Guolin Ke committed
68
  threshold_in_bin_[new_node_idx] = threshold_bin;
Guolin Ke's avatar
Guolin Ke committed
69
  threshold_[new_node_idx] = threshold_double;
70
71
72
  ++num_leaves_;
  return num_leaves_ - 1;
}
Guolin Ke's avatar
Guolin Ke committed
73

74
75
int Tree::SplitCategorical(int leaf, int feature, int real_feature, const uint32_t* threshold_bin, int num_threshold_bin,
                           const uint32_t* threshold, int num_threshold, double left_value, double right_value,
76
77
                           data_size_t left_cnt, data_size_t right_cnt, double left_weight, double right_weight, float gain, MissingType missing_type) {
  Split(leaf, feature, real_feature, left_value, right_value, left_cnt, right_cnt, left_weight, right_weight, gain);
78
79
80
  int new_node_idx = num_leaves_ - 1;
  decision_type_[new_node_idx] = 0;
  SetDecisionType(&decision_type_[new_node_idx], true, kCategoricalMask);
Guolin Ke's avatar
Guolin Ke committed
81
  SetMissingType(&decision_type_[new_node_idx], static_cast<int8_t>(missing_type));
82
83
  threshold_in_bin_[new_node_idx] = num_cat_;
  threshold_[new_node_idx] = num_cat_;
84
  ++num_cat_;
85
86
87
88
89
90
91
92
  cat_boundaries_.push_back(cat_boundaries_.back() + num_threshold);
  for (int i = 0; i < num_threshold; ++i) {
    cat_threshold_.push_back(threshold[i]);
  }
  cat_boundaries_inner_.push_back(cat_boundaries_inner_.back() + num_threshold_bin);
  for (int i = 0; i < num_threshold_bin; ++i) {
    cat_threshold_inner_.push_back(threshold_bin[i]);
  }
Guolin Ke's avatar
Guolin Ke committed
93
94
95
96
  ++num_leaves_;
  return num_leaves_ - 1;
}

Guolin Ke's avatar
Guolin Ke committed
97
98
99
100
101
102
103
104
105
106
107
108
109
110
#define PredictionFun(niter, fidx_in_iter, start_pos, decision_fun, iter_idx, \
                      data_idx)                                               \
  std::vector<std::unique_ptr<BinIterator>> iter((niter));                    \
  for (int i = 0; i < (niter); ++i) {                                         \
    iter[i].reset(data->FeatureIterator((fidx_in_iter)));                     \
    iter[i]->Reset((start_pos));                                              \
  }                                                                           \
  for (data_size_t i = start; i < end; ++i) {                                 \
    int node = 0;                                                             \
    while (node >= 0) {                                                       \
      node = decision_fun(iter[(iter_idx)]->Get((data_idx)), node,            \
                          default_bins[node], max_bins[node]);                \
    }                                                                         \
    score[(data_idx)] += static_cast<double>(leaf_value_[~node]);             \
111
112
  }\

113
114
115
116
117
118
119
120
121
122

#define PredictionFunLinear(niter, fidx_in_iter, start_pos, decision_fun,     \
                            iter_idx, data_idx)                               \
  std::vector<std::unique_ptr<BinIterator>> iter((niter));                    \
  for (int i = 0; i < (niter); ++i) {                                         \
    iter[i].reset(data->FeatureIterator((fidx_in_iter)));                     \
    iter[i]->Reset((start_pos));                                              \
  }                                                                           \
  for (data_size_t i = start; i < end; ++i) {                                 \
    int node = 0;                                                             \
123
124
125
126
127
128
    if (num_leaves_ > 1) {                                                    \
      while (node >= 0) {                                                     \
        node = decision_fun(iter[(iter_idx)]->Get((data_idx)), node,          \
                            default_bins[node], max_bins[node]);              \
      }                                                                       \
      node = ~node;                                                           \
129
    }                                                                         \
130
    double add_score = leaf_const_[node];                                     \
131
    bool nan_found = false;                                                   \
132
133
134
    const double* coeff_ptr = leaf_coeff_[node].data();                       \
    const float** data_ptr = feat_ptr[node].data();                           \
    for (size_t j = 0; j < leaf_features_inner_[node].size(); ++j) {          \
135
136
137
138
139
140
141
142
       float feat_val = data_ptr[j][(data_idx)];                              \
       if (std::isnan(feat_val)) {                                            \
          nan_found = true;                                                   \
          break;                                                              \
       }                                                                      \
       add_score += coeff_ptr[j] * feat_val;                                  \
    }                                                                         \
    if (nan_found) {                                                          \
143
       score[(data_idx)] += leaf_value_[node];                                \
144
145
146
147
148
149
    } else {                                                                  \
      score[(data_idx)] += add_score;                                         \
    }                                                                         \
}\


150
void Tree::AddPredictionToScore(const Dataset* data, data_size_t num_data, double* score) const {
151
  if (!is_linear_ && num_leaves_ <= 1) {
Guolin Ke's avatar
Guolin Ke committed
152
    if (leaf_value_[0] != 0.0f) {
Guolin Ke's avatar
Guolin Ke committed
153
#pragma omp parallel for schedule(static, 512) if (num_data >= 1024)
Guolin Ke's avatar
Guolin Ke committed
154
155
156
157
158
159
      for (data_size_t i = 0; i < num_data; ++i) {
        score[i] += leaf_value_[0];
      }
    }
    return;
  }
Guolin Ke's avatar
Guolin Ke committed
160
161
162
163
164
165
166
167
  std::vector<uint32_t> default_bins(num_leaves_ - 1);
  std::vector<uint32_t> max_bins(num_leaves_ - 1);
  for (int i = 0; i < num_leaves_ - 1; ++i) {
    const int fidx = split_feature_inner_[i];
    auto bin_mapper = data->FeatureBinMapper(fidx);
    default_bins[i] = bin_mapper->GetDefaultBin();
    max_bins[i] = bin_mapper->num_bin() - 1;
  }
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
  if (is_linear_) {
    std::vector<std::vector<const float*>> feat_ptr(num_leaves_);
    for (int leaf_num = 0; leaf_num < num_leaves_; ++leaf_num) {
      for (int feat : leaf_features_inner_[leaf_num]) {
        feat_ptr[leaf_num].push_back(data->raw_index(feat));
      }
    }
    if (num_cat_ > 0) {
      if (data->num_features() > num_leaves_ - 1) {
        Threading::For<data_size_t>(0, num_data, 512, [this, &data, score, &default_bins, &max_bins, &feat_ptr]
        (int, data_size_t start, data_size_t end) {
          PredictionFunLinear(num_leaves_ - 1, split_feature_inner_[i], start, DecisionInner, node, i);
        });
      } else {
        Threading::For<data_size_t>(0, num_data, 512, [this, &data, score, &default_bins, &max_bins, &feat_ptr]
        (int, data_size_t start, data_size_t end) {
          PredictionFunLinear(data->num_features(), i, start, DecisionInner, split_feature_inner_[node], i);
        });
      }
187
    } else {
188
189
190
191
192
193
194
195
196
197
198
      if (data->num_features() > num_leaves_ - 1) {
        Threading::For<data_size_t>(0, num_data, 512, [this, &data, score, &default_bins, &max_bins, &feat_ptr]
        (int, data_size_t start, data_size_t end) {
          PredictionFunLinear(num_leaves_ - 1, split_feature_inner_[i], start, NumericalDecisionInner, node, i);
        });
      } else {
        Threading::For<data_size_t>(0, num_data, 512, [this, &data, score, &default_bins, &max_bins, &feat_ptr]
        (int, data_size_t start, data_size_t end) {
          PredictionFunLinear(data->num_features(), i, start, NumericalDecisionInner, split_feature_inner_[node], i);
        });
      }
199
    }
Guolin Ke's avatar
Guolin Ke committed
200
  } else {
201
202
203
204
205
206
207
208
209
210
211
212
    if (num_cat_ > 0) {
      if (data->num_features() > num_leaves_ - 1) {
        Threading::For<data_size_t>(0, num_data, 512, [this, &data, score, &default_bins, &max_bins]
        (int, data_size_t start, data_size_t end) {
          PredictionFun(num_leaves_ - 1, split_feature_inner_[i], start, DecisionInner, node, i);
        });
      } else {
        Threading::For<data_size_t>(0, num_data, 512, [this, &data, score, &default_bins, &max_bins]
        (int, data_size_t start, data_size_t end) {
          PredictionFun(data->num_features(), i, start, DecisionInner, split_feature_inner_[node], i);
        });
      }
213
    } else {
214
215
216
217
218
219
220
221
222
223
224
      if (data->num_features() > num_leaves_ - 1) {
        Threading::For<data_size_t>(0, num_data, 512, [this, &data, score, &default_bins, &max_bins]
        (int, data_size_t start, data_size_t end) {
          PredictionFun(num_leaves_ - 1, split_feature_inner_[i], start, NumericalDecisionInner, node, i);
        });
      } else {
        Threading::For<data_size_t>(0, num_data, 512, [this, &data, score, &default_bins, &max_bins]
        (int, data_size_t start, data_size_t end) {
          PredictionFun(data->num_features(), i, start, NumericalDecisionInner, split_feature_inner_[node], i);
        });
      }
225
    }
Guolin Ke's avatar
Guolin Ke committed
226
  }
Guolin Ke's avatar
Guolin Ke committed
227
228
}

Guolin Ke's avatar
Guolin Ke committed
229
void Tree::AddPredictionToScore(const Dataset* data,
230
231
232
  const data_size_t* used_data_indices,
  data_size_t num_data, double* score) const {
  if (!is_linear_ && num_leaves_ <= 1) {
Guolin Ke's avatar
Guolin Ke committed
233
    if (leaf_value_[0] != 0.0f) {
Guolin Ke's avatar
Guolin Ke committed
234
#pragma omp parallel for schedule(static, 512) if (num_data >= 1024)
Guolin Ke's avatar
Guolin Ke committed
235
236
237
238
      for (data_size_t i = 0; i < num_data; ++i) {
        score[used_data_indices[i]] += leaf_value_[0];
      }
    }
Guolin Ke's avatar
Guolin Ke committed
239
    return;
Guolin Ke's avatar
Guolin Ke committed
240
  }
Guolin Ke's avatar
Guolin Ke committed
241
242
243
244
245
246
247
248
  std::vector<uint32_t> default_bins(num_leaves_ - 1);
  std::vector<uint32_t> max_bins(num_leaves_ - 1);
  for (int i = 0; i < num_leaves_ - 1; ++i) {
    const int fidx = split_feature_inner_[i];
    auto bin_mapper = data->FeatureBinMapper(fidx);
    default_bins[i] = bin_mapper->GetDefaultBin();
    max_bins[i] = bin_mapper->num_bin() - 1;
  }
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
  if (is_linear_) {
    std::vector<std::vector<const float*>> feat_ptr(num_leaves_);
    for (int leaf_num = 0; leaf_num < num_leaves_; ++leaf_num) {
      for (int feat : leaf_features_inner_[leaf_num]) {
        feat_ptr[leaf_num].push_back(data->raw_index(feat));
      }
    }
    if (num_cat_ > 0) {
      if (data->num_features() > num_leaves_ - 1) {
        Threading::For<data_size_t>(0, num_data, 512, [this, &data, score, used_data_indices, &default_bins, &max_bins, &feat_ptr]
        (int, data_size_t start, data_size_t end) {
          PredictionFunLinear(num_leaves_ - 1, split_feature_inner_[i], used_data_indices[start], DecisionInner,
                              node, used_data_indices[i]);
        });
      } else {
        Threading::For<data_size_t>(0, num_data, 512, [this, &data, score, used_data_indices, &default_bins, &max_bins, &feat_ptr]
        (int, data_size_t start, data_size_t end) {
          PredictionFunLinear(data->num_features(), i, used_data_indices[start], DecisionInner, split_feature_inner_[node], used_data_indices[i]);
        });
      }
269
    } else {
270
271
272
273
274
275
276
277
278
279
280
281
282
      if (data->num_features() > num_leaves_ - 1) {
        Threading::For<data_size_t>(0, num_data, 512, [this, &data, score, used_data_indices, &default_bins, &max_bins, &feat_ptr]
        (int, data_size_t start, data_size_t end) {
          PredictionFunLinear(num_leaves_ - 1, split_feature_inner_[i], used_data_indices[start], NumericalDecisionInner,
                              node, used_data_indices[i]);
        });
      } else {
        Threading::For<data_size_t>(0, num_data, 512, [this, &data, score, used_data_indices, &default_bins, &max_bins, &feat_ptr]
        (int, data_size_t start, data_size_t end) {
          PredictionFunLinear(data->num_features(), i, used_data_indices[start], NumericalDecisionInner,
                              split_feature_inner_[node], used_data_indices[i]);
        });
      }
283
    }
Guolin Ke's avatar
Guolin Ke committed
284
  } else {
285
286
287
288
289
290
291
292
293
294
295
296
    if (num_cat_ > 0) {
      if (data->num_features() > num_leaves_ - 1) {
        Threading::For<data_size_t>(0, num_data, 512, [this, &data, score, used_data_indices, &default_bins, &max_bins]
        (int, data_size_t start, data_size_t end) {
          PredictionFun(num_leaves_ - 1, split_feature_inner_[i], used_data_indices[start], DecisionInner, node, used_data_indices[i]);
        });
      } else {
        Threading::For<data_size_t>(0, num_data, 512, [this, &data, score, used_data_indices, &default_bins, &max_bins]
        (int, data_size_t start, data_size_t end) {
          PredictionFun(data->num_features(), i, used_data_indices[start], DecisionInner, split_feature_inner_[node], used_data_indices[i]);
        });
      }
297
    } else {
298
299
300
301
302
303
304
305
306
307
308
      if (data->num_features() > num_leaves_ - 1) {
        Threading::For<data_size_t>(0, num_data, 512, [this, &data, score, used_data_indices, &default_bins, &max_bins]
        (int, data_size_t start, data_size_t end) {
          PredictionFun(num_leaves_ - 1, split_feature_inner_[i], used_data_indices[start], NumericalDecisionInner, node, used_data_indices[i]);
        });
      } else {
        Threading::For<data_size_t>(0, num_data, 512, [this, &data, score, used_data_indices, &default_bins, &max_bins]
        (int, data_size_t start, data_size_t end) {
          PredictionFun(data->num_features(), i, used_data_indices[start], NumericalDecisionInner, split_feature_inner_[node], used_data_indices[i]);
        });
      }
309
    }
Guolin Ke's avatar
Guolin Ke committed
310
  }
Guolin Ke's avatar
Guolin Ke committed
311
312
}

313
#undef PredictionFun
314
#undef PredictionFunLinear
315

316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
double Tree::GetUpperBoundValue() const {
  double upper_bound = leaf_value_[0];
  for (int i = 1; i < num_leaves_; ++i) {
    if (leaf_value_[i] > upper_bound) {
      upper_bound = leaf_value_[i];
    }
  }
  return upper_bound;
}

double Tree::GetLowerBoundValue() const {
  double lower_bound = leaf_value_[0];
  for (int i = 1; i < num_leaves_; ++i) {
    if (leaf_value_[i] < lower_bound) {
      lower_bound = leaf_value_[i];
    }
  }
  return lower_bound;
}

Guolin Ke's avatar
Guolin Ke committed
336
std::string Tree::ToString() const {
337
  std::stringstream str_buf;
338
339
340
341
342
343
344
345
  Common::C_stringstream(str_buf);

  #if ((defined(sun) || defined(__sun)) && (defined(__SVR4) || defined(__svr4__)))
  using CommonLegacy::ArrayToString;  // Slower & unsafe regarding locale.
  #else
  using CommonC::ArrayToString;
  #endif

346
347
  str_buf << "num_leaves=" << num_leaves_ << '\n';
  str_buf << "num_cat=" << num_cat_ << '\n';
348
  str_buf << "split_feature="
349
    << ArrayToString(split_feature_, num_leaves_ - 1) << '\n';
350
  str_buf << "split_gain="
351
    << ArrayToString(split_gain_, num_leaves_ - 1) << '\n';
352
  str_buf << "threshold="
353
    << ArrayToString<true>(threshold_, num_leaves_ - 1) << '\n';
354
  str_buf << "decision_type="
355
    << ArrayToString(Common::ArrayCast<int8_t, int>(decision_type_), num_leaves_ - 1) << '\n';
356
  str_buf << "left_child="
357
    << ArrayToString(left_child_, num_leaves_ - 1) << '\n';
358
  str_buf << "right_child="
359
    << ArrayToString(right_child_, num_leaves_ - 1) << '\n';
360
  str_buf << "leaf_value="
361
    << ArrayToString<true>(leaf_value_, num_leaves_) << '\n';
362
  str_buf << "leaf_weight="
363
    << ArrayToString<true>(leaf_weight_, num_leaves_) << '\n';
364
  str_buf << "leaf_count="
365
    << ArrayToString(leaf_count_, num_leaves_) << '\n';
366
  str_buf << "internal_value="
367
    << ArrayToString(internal_value_, num_leaves_ - 1) << '\n';
368
  str_buf << "internal_weight="
369
    << ArrayToString(internal_weight_, num_leaves_ - 1) << '\n';
370
  str_buf << "internal_count="
371
    << ArrayToString(internal_count_, num_leaves_ - 1) << '\n';
372
373
  if (num_cat_ > 0) {
    str_buf << "cat_boundaries="
374
      << ArrayToString(cat_boundaries_, num_cat_ + 1) << '\n';
375
    str_buf << "cat_threshold="
376
      << ArrayToString(cat_threshold_, cat_threshold_.size()) << '\n';
377
  }
378
379
380
381
  str_buf << "is_linear=" << is_linear_ << '\n';

  if (is_linear_) {
    str_buf << "leaf_const="
382
      << ArrayToString<true>(leaf_const_, num_leaves_) << '\n';
383
384
    std::vector<int> num_feat(num_leaves_);
    for (int i = 0; i < num_leaves_; ++i) {
385
      num_feat[i] = static_cast<int>(leaf_coeff_[i].size());
386
387
388
389
390
391
392
393
394
395
396
397
398
399
    }
    str_buf << "num_features="
      << ArrayToString(num_feat, num_leaves_) << '\n';
    str_buf << "leaf_features=";
    for (int i = 0; i < num_leaves_; ++i) {
      if (num_feat[i] > 0) {
        str_buf << ArrayToString(leaf_features_[i], leaf_features_[i].size()) << ' ';
      }
      str_buf << ' ';
    }
    str_buf << '\n';
    str_buf << "leaf_coeff=";
    for (int i = 0; i < num_leaves_; ++i) {
      if (num_feat[i] > 0) {
400
        str_buf << ArrayToString<true>(leaf_coeff_[i], leaf_coeff_[i].size()) << ' ';
401
402
403
404
405
      }
      str_buf << ' ';
    }
    str_buf << '\n';
  }
406
407
  str_buf << "shrinkage=" << shrinkage_ << '\n';
  str_buf << '\n';
408

409
  return str_buf.str();
Guolin Ke's avatar
Guolin Ke committed
410
411
}

Guolin Ke's avatar
Guolin Ke committed
412
std::string Tree::ToJSON() const {
413
  std::stringstream str_buf;
414
  Common::C_stringstream(str_buf);
415
  str_buf << std::setprecision(std::numeric_limits<double>::digits10 + 2);
416
417
418
  str_buf << "\"num_leaves\":" << num_leaves_ << "," << '\n';
  str_buf << "\"num_cat\":" << num_cat_ << "," << '\n';
  str_buf << "\"shrinkage\":" << shrinkage_ << "," << '\n';
wxchan's avatar
wxchan committed
419
  if (num_leaves_ == 1) {
420
421
422
423
424
425
    if (is_linear_) {
      str_buf << "\"tree_structure\":{" << "\"leaf_value\":" << leaf_value_[0] << ", " << "\n";
      str_buf << LinearModelToJSON(0) << "}" << "\n";
    } else {
      str_buf << "\"tree_structure\":{" << "\"leaf_value\":" << leaf_value_[0] << "}" << '\n';
    }
wxchan's avatar
wxchan committed
426
  } else {
427
    str_buf << "\"tree_structure\":" << NodeToJSON(0) << '\n';
wxchan's avatar
wxchan committed
428
  }
429
430
  return str_buf.str();
}
wxchan's avatar
wxchan committed
431

432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
std::string Tree::LinearModelToJSON(int index) const {
  std::stringstream str_buf;
  Common::C_stringstream(str_buf);
  str_buf << std::setprecision(std::numeric_limits<double>::digits10 + 2);
  str_buf << "\"leaf_const\":" << leaf_const_[index] << "," << "\n";
  int num_features = static_cast<int>(leaf_features_[index].size());
  if (num_features > 0) {
    str_buf << "\"leaf_features\":[";
    for (int i = 0; i < num_features - 1; ++i) {
      str_buf << leaf_features_[index][i] << ", ";
    }
    str_buf << leaf_features_[index][num_features - 1] << "]" << ", " << "\n";
    str_buf << "\"leaf_coeff\":[";
    for (int i = 0; i < num_features - 1; ++i) {
      str_buf << leaf_coeff_[index][i] << ", ";
    }
    str_buf << leaf_coeff_[index][num_features - 1] << "]" << "\n";
  } else {
    str_buf << "\"leaf_features\":[],\n";
    str_buf << "\"leaf_coeff\":[]\n";
  }
453
  return str_buf.str();
wxchan's avatar
wxchan committed
454
455
}

Guolin Ke's avatar
Guolin Ke committed
456
std::string Tree::NodeToJSON(int index) const {
457
  std::stringstream str_buf;
458
  Common::C_stringstream(str_buf);
459
  str_buf << std::setprecision(std::numeric_limits<double>::digits10 + 2);
wxchan's avatar
wxchan committed
460
461
  if (index >= 0) {
    // non-leaf
462
463
464
    str_buf << "{" << '\n';
    str_buf << "\"split_index\":" << index << "," << '\n';
    str_buf << "\"split_feature\":" << split_feature_[index] << "," << '\n';
Guolin Ke's avatar
Guolin Ke committed
465
    str_buf << "\"split_gain\":" << Common::AvoidInf(split_gain_[index]) << "," << '\n';
466
    if (GetDecisionType(decision_type_[index], kCategoricalMask)) {
467
468
469
470
471
472
473
474
475
476
477
      int cat_idx = static_cast<int>(threshold_[index]);
      std::vector<int> cats;
      for (int i = cat_boundaries_[cat_idx]; i < cat_boundaries_[cat_idx + 1]; ++i) {
        for (int j = 0; j < 32; ++j) {
          int cat = (i - cat_boundaries_[cat_idx]) * 32 + j;
          if (Common::FindInBitset(cat_threshold_.data() + cat_boundaries_[cat_idx],
                                   cat_boundaries_[cat_idx + 1] - cat_boundaries_[cat_idx], cat)) {
            cats.push_back(cat);
          }
        }
      }
478
      str_buf << "\"threshold\":\"" << CommonC::Join(cats, "||") << "\"," << '\n';
479
      str_buf << "\"decision_type\":\"==\"," << '\n';
480
    } else {
481
482
      str_buf << "\"threshold\":" << Common::AvoidInf(threshold_[index]) << "," << '\n';
      str_buf << "\"decision_type\":\"<=\"," << '\n';
483
484
    }
    if (GetDecisionType(decision_type_[index], kDefaultLeftMask)) {
485
      str_buf << "\"default_left\":true," << '\n';
486
    } else {
487
      str_buf << "\"default_left\":false," << '\n';
488
489
    }
    uint8_t missing_type = GetMissingType(decision_type_[index]);
490
    if (missing_type == MissingType::None) {
491
      str_buf << "\"missing_type\":\"None\"," << '\n';
492
    } else if (missing_type == MissingType::Zero) {
493
      str_buf << "\"missing_type\":\"Zero\"," << '\n';
494
    } else {
495
      str_buf << "\"missing_type\":\"NaN\"," << '\n';
496
    }
497
    str_buf << "\"internal_value\":" << internal_value_[index] << "," << '\n';
498
    str_buf << "\"internal_weight\":" << internal_weight_[index] << "," << '\n';
499
500
501
    str_buf << "\"internal_count\":" << internal_count_[index] << "," << '\n';
    str_buf << "\"left_child\":" << NodeToJSON(left_child_[index]) << "," << '\n';
    str_buf << "\"right_child\":" << NodeToJSON(right_child_[index]) << '\n';
502
    str_buf << "}";
wxchan's avatar
wxchan committed
503
504
505
  } else {
    // leaf
    index = ~index;
506
507
508
    str_buf << "{" << '\n';
    str_buf << "\"leaf_index\":" << index << "," << '\n';
    str_buf << "\"leaf_value\":" << leaf_value_[index] << "," << '\n';
509
    str_buf << "\"leaf_weight\":" << leaf_weight_[index] << "," << '\n';
510
511
512
513
514
515
    if (is_linear_) {
      str_buf << "\"leaf_count\":" << leaf_count_[index] << "," << '\n';
      str_buf << LinearModelToJSON(index);
    } else {
      str_buf << "\"leaf_count\":" << leaf_count_[index] << '\n';
    }
516
    str_buf << "}";
wxchan's avatar
wxchan committed
517
  }
518
  return str_buf.str();
wxchan's avatar
wxchan committed
519
520
}

Guolin Ke's avatar
Guolin Ke committed
521
522
std::string Tree::NumericalDecisionIfElse(int node) const {
  std::stringstream str_buf;
523
  Common::C_stringstream(str_buf);
Guolin Ke's avatar
Guolin Ke committed
524
525
  uint8_t missing_type = GetMissingType(decision_type_[node]);
  bool default_left = GetDecisionType(decision_type_[node], kDefaultLeftMask);
Nikita Titov's avatar
Nikita Titov committed
526
  if (missing_type == MissingType::None
527
      || (missing_type == MissingType::Zero && default_left && kZeroThreshold < threshold_[node])) {
Guolin Ke's avatar
Guolin Ke committed
528
    str_buf << "if (fval <= " << threshold_[node] << ") {";
529
  } else if (missing_type == MissingType::Zero) {
Guolin Ke's avatar
Guolin Ke committed
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
    if (default_left) {
      str_buf << "if (fval <= " << threshold_[node] << " || Tree::IsZero(fval)" << " || std::isnan(fval)) {";
    } else {
      str_buf << "if (fval <= " << threshold_[node] << " && !Tree::IsZero(fval)" << " && !std::isnan(fval)) {";
    }
  } else {
    if (default_left) {
      str_buf << "if (fval <= " << threshold_[node] << " || std::isnan(fval)) {";
    } else {
      str_buf << "if (fval <= " << threshold_[node] << " && !std::isnan(fval)) {";
    }
  }
  return str_buf.str();
}

std::string Tree::CategoricalDecisionIfElse(int node) const {
  uint8_t missing_type = GetMissingType(decision_type_[node]);
  std::stringstream str_buf;
548
  Common::C_stringstream(str_buf);
549
  if (missing_type == MissingType::NaN) {
Guolin Ke's avatar
Guolin Ke committed
550
551
552
553
    str_buf << "if (std::isnan(fval)) { int_fval = -1; } else { int_fval = static_cast<int>(fval); }";
  } else {
    str_buf << "if (std::isnan(fval)) { int_fval = 0; } else { int_fval = static_cast<int>(fval); }";
  }
554
  int cat_idx = static_cast<int>(threshold_[node]);
555
556
557
558
  str_buf << "if (int_fval >= 0 && int_fval < 32 * (";
  str_buf << cat_boundaries_[cat_idx + 1] - cat_boundaries_[cat_idx];
  str_buf << ") && (((cat_threshold[" << cat_boundaries_[cat_idx];
  str_buf << " + int_fval / 32] >> (int_fval & 31)) & 1))) {";
Guolin Ke's avatar
Guolin Ke committed
559
560
561
  return str_buf.str();
}

Guolin Ke's avatar
Guolin Ke committed
562
std::string Tree::ToIfElse(int index, bool predict_leaf_index) const {
563
  std::stringstream str_buf;
564
  Common::C_stringstream(str_buf);
565
  str_buf << "double PredictTree" << index;
Guolin Ke's avatar
Guolin Ke committed
566
  if (predict_leaf_index) {
567
568
569
    str_buf << "Leaf";
  }
  str_buf << "(const double* arr) { ";
570
571
  if (num_leaves_ <= 1) {
    str_buf << "return " << leaf_value_[0] << ";";
572
  } else {
573
574
575
576
577
578
579
580
    str_buf << "const std::vector<uint32_t> cat_threshold = {";
    for (size_t i = 0; i < cat_threshold_.size(); ++i) {
      if (i != 0) {
        str_buf << ",";
      }
      str_buf << cat_threshold_[i];
    }
    str_buf << "};";
581
582
583
584
585
    // use this for the missing value conversion
    str_buf << "double fval = 0.0f; ";
    if (num_cat_ > 0) {
      str_buf << "int int_fval = 0; ";
    }
Guolin Ke's avatar
Guolin Ke committed
586
    str_buf << NodeToIfElse(0, predict_leaf_index);
587
  }
588
  str_buf << " }" << '\n';
589

590
  // Predict func by Map to ifelse
591
  str_buf << "double PredictTree" << index;
Guolin Ke's avatar
Guolin Ke committed
592
  if (predict_leaf_index) {
Guolin Ke's avatar
Guolin Ke committed
593
594
595
    str_buf << "LeafByMap";
  } else {
    str_buf << "ByMap";
596
597
598
  }
  str_buf << "(const std::unordered_map<int, double>& arr) { ";
  if (num_leaves_ <= 1) {
Guolin Ke's avatar
Guolin Ke committed
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
    str_buf << "return " << leaf_value_[0] << ";";
  } else {
    str_buf << "const std::vector<uint32_t> cat_threshold = {";
    for (size_t i = 0; i < cat_threshold_.size(); ++i) {
      if (i != 0) {
        str_buf << ",";
      }
      str_buf << cat_threshold_[i];
    }
    str_buf << "};";
    // use this for the missing value conversion
    str_buf << "double fval = 0.0f; ";
    if (num_cat_ > 0) {
      str_buf << "int int_fval = 0; ";
    }
Guolin Ke's avatar
Guolin Ke committed
614
    str_buf << NodeToIfElseByMap(0, predict_leaf_index);
615
  }
616
  str_buf << " }" << '\n';
617

618
619
620
  return str_buf.str();
}

Guolin Ke's avatar
Guolin Ke committed
621
std::string Tree::NodeToIfElse(int index, bool predict_leaf_index) const {
622
  std::stringstream str_buf;
623
  Common::C_stringstream(str_buf);
624
625
626
  str_buf << std::setprecision(std::numeric_limits<double>::digits10 + 2);
  if (index >= 0) {
    // non-leaf
627
    str_buf << "fval = arr[" << split_feature_[index] << "];";
Guolin Ke's avatar
Guolin Ke committed
628
    if (GetDecisionType(decision_type_[index], kCategoricalMask) == 0) {
629
      str_buf << NumericalDecisionIfElse(index);
630
    } else {
631
      str_buf << CategoricalDecisionIfElse(index);
632
633
    }
    // left subtree
Guolin Ke's avatar
Guolin Ke committed
634
    str_buf << NodeToIfElse(left_child_[index], predict_leaf_index);
635
636
    str_buf << " } else { ";
    // right subtree
Guolin Ke's avatar
Guolin Ke committed
637
    str_buf << NodeToIfElse(right_child_[index], predict_leaf_index);
638
639
640
641
    str_buf << " }";
  } else {
    // leaf
    str_buf << "return ";
Guolin Ke's avatar
Guolin Ke committed
642
    if (predict_leaf_index) {
643
644
645
646
647
648
649
650
651
652
      str_buf << ~index;
    } else {
      str_buf << leaf_value_[~index];
    }
    str_buf << ";";
  }

  return str_buf.str();
}

Guolin Ke's avatar
Guolin Ke committed
653
std::string Tree::NodeToIfElseByMap(int index, bool predict_leaf_index) const {
654
  std::stringstream str_buf;
655
  Common::C_stringstream(str_buf);
656
657
658
659
660
661
662
663
664
665
  str_buf << std::setprecision(std::numeric_limits<double>::digits10 + 2);
  if (index >= 0) {
    // non-leaf
    str_buf << "fval = arr.count(" << split_feature_[index] << ") > 0 ? arr.at(" << split_feature_[index] << ") : 0.0f;";
    if (GetDecisionType(decision_type_[index], kCategoricalMask) == 0) {
      str_buf << NumericalDecisionIfElse(index);
    } else {
      str_buf << CategoricalDecisionIfElse(index);
    }
    // left subtree
Guolin Ke's avatar
Guolin Ke committed
666
    str_buf << NodeToIfElseByMap(left_child_[index], predict_leaf_index);
667
668
    str_buf << " } else { ";
    // right subtree
Guolin Ke's avatar
Guolin Ke committed
669
    str_buf << NodeToIfElseByMap(right_child_[index], predict_leaf_index);
670
671
672
673
    str_buf << " }";
  } else {
    // leaf
    str_buf << "return ";
Guolin Ke's avatar
Guolin Ke committed
674
    if (predict_leaf_index) {
675
676
677
678
679
680
681
682
683
684
      str_buf << ~index;
    } else {
      str_buf << leaf_value_[~index];
    }
    str_buf << ";";
  }

  return str_buf.str();
}

685
686
Tree::Tree(const char* str, size_t* used_len) {
  auto p = str;
Guolin Ke's avatar
Guolin Ke committed
687
  std::unordered_map<std::string, std::string> key_vals;
688
  const int max_num_line = 22;
689
690
691
692
  int read_line = 0;
  while (read_line < max_num_line) {
    if (*p == '\r' || *p == '\n') break;
    auto start = p;
693
    while (*p != '=') ++p;
694
695
696
697
698
699
700
701
702
703
704
    std::string key(start, p - start);
    ++p;
    start = p;
    while (*p != '\r' && *p != '\n') ++p;
    key_vals[key] = std::string(start, p - start);
    ++read_line;
    if (*p == '\r') ++p;
    if (*p == '\n') ++p;
  }
  *used_len = p - str;

705
  if (key_vals.count("num_leaves") <= 0) {
706
    Log::Fatal("Tree model should contain num_leaves field");
Guolin Ke's avatar
Guolin Ke committed
707
708
709
710
  }

  Common::Atoi(key_vals["num_leaves"].c_str(), &num_leaves_);

711
  if (key_vals.count("num_cat") <= 0) {
712
    Log::Fatal("Tree model should contain num_cat field");
713
714
715
716
  }

  Common::Atoi(key_vals["num_cat"].c_str(), &num_cat_);

717
  if (key_vals.count("leaf_value")) {
718
    leaf_value_ = CommonC::StringToArray<double>(key_vals["leaf_value"], num_leaves_);
719
720
721
  } else {
    Log::Fatal("Tree model string format error, should contain leaf_value field");
  }
722

Guolin Ke's avatar
Guolin Ke committed
723
  if (key_vals.count("shrinkage")) {
724
    CommonC::Atof(key_vals["shrinkage"].c_str(), &shrinkage_);
Guolin Ke's avatar
Guolin Ke committed
725
726
727
  } else {
    shrinkage_ = 1.0f;
  }
728

729
730
731
732
  if (key_vals.count("is_linear")) {
    int is_linear_int;
    Common::Atoi(key_vals["is_linear"].c_str(), &is_linear_int);
    is_linear_ = static_cast<bool>(is_linear_int);
733
734
  } else {
    is_linear_ = false;
735
736
  }

737
738
739
  if ((num_leaves_ <= 1) && !is_linear_) {
    return;
  }
740

Guolin Ke's avatar
Guolin Ke committed
741
  if (key_vals.count("left_child")) {
742
    left_child_ = CommonC::StringToArrayFast<int>(key_vals["left_child"], num_leaves_ - 1);
Guolin Ke's avatar
Guolin Ke committed
743
744
  } else {
    Log::Fatal("Tree model string format error, should contain left_child field");
745
746
  }

Guolin Ke's avatar
Guolin Ke committed
747
  if (key_vals.count("right_child")) {
748
    right_child_ = CommonC::StringToArrayFast<int>(key_vals["right_child"], num_leaves_ - 1);
Guolin Ke's avatar
Guolin Ke committed
749
750
751
752
753
  } else {
    Log::Fatal("Tree model string format error, should contain right_child field");
  }

  if (key_vals.count("split_feature")) {
754
    split_feature_ = CommonC::StringToArrayFast<int>(key_vals["split_feature"], num_leaves_ - 1);
Guolin Ke's avatar
Guolin Ke committed
755
756
757
758
759
  } else {
    Log::Fatal("Tree model string format error, should contain split_feature field");
  }

  if (key_vals.count("threshold")) {
760
    threshold_ = CommonC::StringToArray<double>(key_vals["threshold"], num_leaves_ - 1);
Guolin Ke's avatar
Guolin Ke committed
761
762
763
764
765
  } else {
    Log::Fatal("Tree model string format error, should contain threshold field");
  }

  if (key_vals.count("split_gain")) {
766
    split_gain_ = CommonC::StringToArrayFast<float>(key_vals["split_gain"], num_leaves_ - 1);
Guolin Ke's avatar
Guolin Ke committed
767
768
769
770
771
  } else {
    split_gain_.resize(num_leaves_ - 1);
  }

  if (key_vals.count("internal_count")) {
772
    internal_count_ = CommonC::StringToArrayFast<int>(key_vals["internal_count"], num_leaves_ - 1);
Guolin Ke's avatar
Guolin Ke committed
773
774
775
776
777
  } else {
    internal_count_.resize(num_leaves_ - 1);
  }

  if (key_vals.count("internal_value")) {
778
    internal_value_ = CommonC::StringToArrayFast<double>(key_vals["internal_value"], num_leaves_ - 1);
Guolin Ke's avatar
Guolin Ke committed
779
780
781
782
  } else {
    internal_value_.resize(num_leaves_ - 1);
  }

783
  if (key_vals.count("internal_weight")) {
784
    internal_weight_ = CommonC::StringToArrayFast<double>(key_vals["internal_weight"], num_leaves_ - 1);
785
  } else {
786
787
788
789
    internal_weight_.resize(num_leaves_ - 1);
  }

  if (key_vals.count("leaf_weight")) {
790
    leaf_weight_ = CommonC::StringToArray<double>(key_vals["leaf_weight"], num_leaves_);
791
  } else {
792
793
794
    leaf_weight_.resize(num_leaves_);
  }

Guolin Ke's avatar
Guolin Ke committed
795
  if (key_vals.count("leaf_count")) {
796
    leaf_count_ = CommonC::StringToArrayFast<int>(key_vals["leaf_count"], num_leaves_);
Guolin Ke's avatar
Guolin Ke committed
797
798
799
800
801
  } else {
    leaf_count_.resize(num_leaves_);
  }

  if (key_vals.count("decision_type")) {
802
    decision_type_ = CommonC::StringToArrayFast<int8_t>(key_vals["decision_type"], num_leaves_ - 1);
Guolin Ke's avatar
Guolin Ke committed
803
804
805
806
  } else {
    decision_type_ = std::vector<int8_t>(num_leaves_ - 1, 0);
  }

807
808
  if (is_linear_) {
    if (key_vals.count("leaf_const")) {
809
      leaf_const_ = Common::StringToArray<double>(key_vals["leaf_const"], num_leaves_);
810
811
812
813
814
815
816
817
818
819
820
821
    } else {
      leaf_const_.resize(num_leaves_);
    }
    std::vector<int> num_feat;
    if (key_vals.count("num_features")) {
      num_feat = Common::StringToArrayFast<int>(key_vals["num_features"], num_leaves_);
    }
    leaf_coeff_.resize(num_leaves_);
    leaf_features_.resize(num_leaves_);
    leaf_features_inner_.resize(num_leaves_);
    if (num_feat.size() > 0) {
      int total_num_feat = 0;
822
823
824
      for (size_t i = 0; i < num_feat.size(); ++i) {
        total_num_feat += num_feat[i];
      }
825
826
827
828
829
830
      std::vector<int> all_leaf_features;
      if (key_vals.count("leaf_features")) {
        all_leaf_features = Common::StringToArrayFast<int>(key_vals["leaf_features"], total_num_feat);
      }
      std::vector<double> all_leaf_coeff;
      if (key_vals.count("leaf_coeff")) {
831
        all_leaf_coeff = Common::StringToArray<double>(key_vals["leaf_coeff"], total_num_feat);
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
      }
      int sum_num_feat = 0;
      for (int i = 0; i < num_leaves_; ++i) {
        if (num_feat[i] > 0) {
          if (key_vals.count("leaf_features"))  {
            leaf_features_[i].assign(all_leaf_features.begin() + sum_num_feat, all_leaf_features.begin() + sum_num_feat + num_feat[i]);
          }
          if (key_vals.count("leaf_coeff")) {
            leaf_coeff_[i].assign(all_leaf_coeff.begin() + sum_num_feat, all_leaf_coeff.begin() + sum_num_feat + num_feat[i]);
          }
        }
        sum_num_feat += num_feat[i];
      }
    }
  }

848
849
  if (num_cat_ > 0) {
    if (key_vals.count("cat_boundaries")) {
850
      cat_boundaries_ = CommonC::StringToArrayFast<int>(key_vals["cat_boundaries"], num_cat_ + 1);
851
852
853
854
855
    } else {
      Log::Fatal("Tree model should contain cat_boundaries field.");
    }

    if (key_vals.count("cat_threshold")) {
856
      cat_threshold_ = CommonC::StringToArrayFast<uint32_t>(key_vals["cat_threshold"], cat_boundaries_.back());
857
    } else {
858
      Log::Fatal("Tree model should contain cat_threshold field");
859
860
    }
  }
861
  max_depth_ = -1;
Guolin Ke's avatar
Guolin Ke committed
862
863
}

Guolin Ke's avatar
Guolin Ke committed
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
void Tree::ExtendPath(PathElement *unique_path, int unique_depth,
                      double zero_fraction, double one_fraction, int feature_index) {
  unique_path[unique_depth].feature_index = feature_index;
  unique_path[unique_depth].zero_fraction = zero_fraction;
  unique_path[unique_depth].one_fraction = one_fraction;
  unique_path[unique_depth].pweight = (unique_depth == 0 ? 1 : 0);
  for (int i = unique_depth - 1; i >= 0; i--) {
    unique_path[i + 1].pweight += one_fraction*unique_path[i].pweight*(i + 1)
      / static_cast<double>(unique_depth + 1);
    unique_path[i].pweight = zero_fraction*unique_path[i].pweight*(unique_depth - i)
      / static_cast<double>(unique_depth + 1);
  }
}

void Tree::UnwindPath(PathElement *unique_path, int unique_depth, int path_index) {
  const double one_fraction = unique_path[path_index].one_fraction;
  const double zero_fraction = unique_path[path_index].zero_fraction;
  double next_one_portion = unique_path[unique_depth].pweight;

  for (int i = unique_depth - 1; i >= 0; --i) {
    if (one_fraction != 0) {
      const double tmp = unique_path[i].pweight;
      unique_path[i].pweight = next_one_portion*(unique_depth + 1)
        / static_cast<double>((i + 1)*one_fraction);
      next_one_portion = tmp - unique_path[i].pweight*zero_fraction*(unique_depth - i)
        / static_cast<double>(unique_depth + 1);
    } else {
      unique_path[i].pweight = (unique_path[i].pweight*(unique_depth + 1))
        / static_cast<double>(zero_fraction*(unique_depth - i));
    }
  }

  for (int i = path_index; i < unique_depth; ++i) {
    unique_path[i].feature_index = unique_path[i + 1].feature_index;
    unique_path[i].zero_fraction = unique_path[i + 1].zero_fraction;
    unique_path[i].one_fraction = unique_path[i + 1].one_fraction;
  }
}

double Tree::UnwoundPathSum(const PathElement *unique_path, int unique_depth, int path_index) {
  const double one_fraction = unique_path[path_index].one_fraction;
  const double zero_fraction = unique_path[path_index].zero_fraction;
  double next_one_portion = unique_path[unique_depth].pweight;
  double total = 0;
  for (int i = unique_depth - 1; i >= 0; --i) {
    if (one_fraction != 0) {
      const double tmp = next_one_portion*(unique_depth + 1)
        / static_cast<double>((i + 1)*one_fraction);
      total += tmp;
      next_one_portion = unique_path[i].pweight - tmp*zero_fraction*((unique_depth - i)
                                                                     / static_cast<double>(unique_depth + 1));
    } else {
      total += (unique_path[i].pweight / zero_fraction) / ((unique_depth - i)
                                                           / static_cast<double>(unique_depth + 1));
    }
  }
  return total;
}

// recursive computation of SHAP values for a decision tree
void Tree::TreeSHAP(const double *feature_values, double *phi,
                    int node, int unique_depth,
                    PathElement *parent_unique_path, double parent_zero_fraction,
                    double parent_one_fraction, int parent_feature_index) const {
  // extend the unique path
929
  PathElement* unique_path = parent_unique_path + unique_depth;
930
931
932
  if (unique_depth > 0) {
    std::copy(parent_unique_path, parent_unique_path + unique_depth, unique_path);
  }
Guolin Ke's avatar
Guolin Ke committed
933
934
935
936
937
938
939
940
941
942
943
944
945
  ExtendPath(unique_path, unique_depth, parent_zero_fraction,
             parent_one_fraction, parent_feature_index);

  // leaf node
  if (node < 0) {
    for (int i = 1; i <= unique_depth; ++i) {
      const double w = UnwoundPathSum(unique_path, unique_depth, i);
      const PathElement &el = unique_path[i];
      phi[el.feature_index] += w*(el.one_fraction - el.zero_fraction)*leaf_value_[~node];
    }

    // internal node
  } else {
946
    const int hot_index = Decision(feature_values[split_feature_[node]], node);
Guolin Ke's avatar
Guolin Ke committed
947
948
949
950
951
952
953
954
955
956
957
    const int cold_index = (hot_index == left_child_[node] ? right_child_[node] : left_child_[node]);
    const double w = data_count(node);
    const double hot_zero_fraction = data_count(hot_index) / w;
    const double cold_zero_fraction = data_count(cold_index) / w;
    double incoming_zero_fraction = 1;
    double incoming_one_fraction = 1;

    // see if we have already split on this feature,
    // if so we undo that split so we can redo it for this node
    int path_index = 0;
    for (; path_index <= unique_depth; ++path_index) {
958
      if (unique_path[path_index].feature_index == split_feature_[node]) break;
Guolin Ke's avatar
Guolin Ke committed
959
960
961
962
963
964
965
966
967
    }
    if (path_index != unique_depth + 1) {
      incoming_zero_fraction = unique_path[path_index].zero_fraction;
      incoming_one_fraction = unique_path[path_index].one_fraction;
      UnwindPath(unique_path, unique_depth, path_index);
      unique_depth -= 1;
    }

    TreeSHAP(feature_values, phi, hot_index, unique_depth + 1, unique_path,
968
             hot_zero_fraction*incoming_zero_fraction, incoming_one_fraction, split_feature_[node]);
Guolin Ke's avatar
Guolin Ke committed
969
970

    TreeSHAP(feature_values, phi, cold_index, unique_depth + 1, unique_path,
971
             cold_zero_fraction*incoming_zero_fraction, 0, split_feature_[node]);
Guolin Ke's avatar
Guolin Ke committed
972
973
974
  }
}

975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
// recursive sparse computation of SHAP values for a decision tree
void Tree::TreeSHAPByMap(const std::unordered_map<int, double>& feature_values, std::unordered_map<int, double>* phi,
                         int node, int unique_depth,
                         PathElement *parent_unique_path, double parent_zero_fraction,
                         double parent_one_fraction, int parent_feature_index) const {
  // extend the unique path
  PathElement* unique_path = parent_unique_path + unique_depth;
  if (unique_depth > 0) {
    std::copy(parent_unique_path, parent_unique_path + unique_depth, unique_path);
  }
  ExtendPath(unique_path, unique_depth, parent_zero_fraction,
             parent_one_fraction, parent_feature_index);

  // leaf node
  if (node < 0) {
    for (int i = 1; i <= unique_depth; ++i) {
      const double w = UnwoundPathSum(unique_path, unique_depth, i);
      const PathElement &el = unique_path[i];
      (*phi)[el.feature_index] += w*(el.one_fraction - el.zero_fraction)*leaf_value_[~node];
    }

  // internal node
  } else {
    const int hot_index = Decision(feature_values.count(split_feature_[node]) > 0 ? feature_values.at(split_feature_[node]) : 0.0f, node);
    const int cold_index = (hot_index == left_child_[node] ? right_child_[node] : left_child_[node]);
    const double w = data_count(node);
    const double hot_zero_fraction = data_count(hot_index) / w;
    const double cold_zero_fraction = data_count(cold_index) / w;
    double incoming_zero_fraction = 1;
    double incoming_one_fraction = 1;

    // see if we have already split on this feature,
    // if so we undo that split so we can redo it for this node
    int path_index = 0;
    for (; path_index <= unique_depth; ++path_index) {
      if (unique_path[path_index].feature_index == split_feature_[node]) break;
    }
    if (path_index != unique_depth + 1) {
      incoming_zero_fraction = unique_path[path_index].zero_fraction;
      incoming_one_fraction = unique_path[path_index].one_fraction;
      UnwindPath(unique_path, unique_depth, path_index);
      unique_depth -= 1;
    }

    TreeSHAPByMap(feature_values, phi, hot_index, unique_depth + 1, unique_path,
                  hot_zero_fraction*incoming_zero_fraction, incoming_one_fraction, split_feature_[node]);

    TreeSHAPByMap(feature_values, phi, cold_index, unique_depth + 1, unique_path,
                  cold_zero_fraction*incoming_zero_fraction, 0, split_feature_[node]);
  }
}

1027
1028
1029
1030
1031
double Tree::ExpectedValue() const {
  if (num_leaves_ == 1) return LeafOutput(0);
  const double total_count = internal_count_[0];
  double exp_value = 0.0;
  for (int i = 0; i < num_leaves(); ++i) {
Guolin Ke's avatar
Guolin Ke committed
1032
    exp_value += (leaf_count_[i] / total_count)*LeafOutput(i);
Guolin Ke's avatar
Guolin Ke committed
1033
  }
1034
  return exp_value;
Guolin Ke's avatar
Guolin Ke committed
1035
1036
}

1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
void Tree::RecomputeMaxDepth() {
  if (num_leaves_ == 1) {
    max_depth_ = 0;
  } else {
    if (leaf_depth_.size() == 0) {
      RecomputeLeafDepths(0, 0);
    }
    max_depth_ = leaf_depth_[0];
    for (int i = 1; i < num_leaves(); ++i) {
      if (max_depth_ < leaf_depth_[i]) max_depth_ = leaf_depth_[i];
    }
Guolin Ke's avatar
Guolin Ke committed
1048
1049
1050
  }
}

Guolin Ke's avatar
Guolin Ke committed
1051
}  // namespace LightGBM