config.h 42.7 KB
Newer Older
1
2
3
4
5
6
7
/*!
 * Copyright (c) 2016 Microsoft Corporation. All rights reserved.
 * Licensed under the MIT License. See LICENSE file in the project root for license information.
 *
 * \note
 * desc and descl2 fields must be written in reStructuredText format
 */
Guolin Ke's avatar
Guolin Ke committed
8
9
10
#ifndef LIGHTGBM_CONFIG_H_
#define LIGHTGBM_CONFIG_H_

11
12
#include <LightGBM/export.h>
#include <LightGBM/meta.h>
Guolin Ke's avatar
Guolin Ke committed
13
14
15
16
17
#include <LightGBM/utils/common.h>
#include <LightGBM/utils/log.h>

#include <string>
#include <algorithm>
Guolin Ke's avatar
Guolin Ke committed
18
#include <memory>
19
20
21
#include <unordered_map>
#include <unordered_set>
#include <vector>
Guolin Ke's avatar
Guolin Ke committed
22
23
24

namespace LightGBM {

Guolin Ke's avatar
Guolin Ke committed
25
26
27
28
/*! \brief Types of tasks */
enum TaskType {
  kTrain, kPredict, kConvertModel, KRefitTree
};
29
const int kDefaultNumLeaves = 31;
Guolin Ke's avatar
Guolin Ke committed
30

Guolin Ke's avatar
Guolin Ke committed
31
struct Config {
Nikita Titov's avatar
Nikita Titov committed
32
 public:
Guolin Ke's avatar
Guolin Ke committed
33
  std::string ToString() const;
Guolin Ke's avatar
Guolin Ke committed
34
35
36
37
  /*!
  * \brief Get string value by specific name of key
  * \param params Store the key and value for params
  * \param name Name of key
Hui Xue's avatar
Hui Xue committed
38
  * \param out Value will assign to out if key exists
Guolin Ke's avatar
Guolin Ke committed
39
40
  * \return True if key exists
  */
Guolin Ke's avatar
Guolin Ke committed
41
  inline static bool GetString(
Guolin Ke's avatar
Guolin Ke committed
42
43
44
45
46
47
48
    const std::unordered_map<std::string, std::string>& params,
    const std::string& name, std::string* out);

  /*!
  * \brief Get int value by specific name of key
  * \param params Store the key and value for params
  * \param name Name of key
Hui Xue's avatar
Hui Xue committed
49
  * \param out Value will assign to out if key exists
Guolin Ke's avatar
Guolin Ke committed
50
51
  * \return True if key exists
  */
Guolin Ke's avatar
Guolin Ke committed
52
  inline static bool GetInt(
Guolin Ke's avatar
Guolin Ke committed
53
54
55
56
    const std::unordered_map<std::string, std::string>& params,
    const std::string& name, int* out);

  /*!
57
  * \brief Get double value by specific name of key
Guolin Ke's avatar
Guolin Ke committed
58
59
  * \param params Store the key and value for params
  * \param name Name of key
Hui Xue's avatar
Hui Xue committed
60
  * \param out Value will assign to out if key exists
Guolin Ke's avatar
Guolin Ke committed
61
62
  * \return True if key exists
  */
Guolin Ke's avatar
Guolin Ke committed
63
  inline static bool GetDouble(
Guolin Ke's avatar
Guolin Ke committed
64
    const std::unordered_map<std::string, std::string>& params,
65
    const std::string& name, double* out);
Guolin Ke's avatar
Guolin Ke committed
66
67
68
69
70

  /*!
  * \brief Get bool value by specific name of key
  * \param params Store the key and value for params
  * \param name Name of key
Hui Xue's avatar
Hui Xue committed
71
  * \param out Value will assign to out if key exists
Guolin Ke's avatar
Guolin Ke committed
72
73
  * \return True if key exists
  */
Guolin Ke's avatar
Guolin Ke committed
74
  inline static bool GetBool(
Guolin Ke's avatar
Guolin Ke committed
75
76
    const std::unordered_map<std::string, std::string>& params,
    const std::string& name, bool* out);
77

wxchan's avatar
wxchan committed
78
  static void KV2Map(std::unordered_map<std::string, std::string>& params, const char* kv);
79
  static std::unordered_map<std::string, std::string> Str2Map(const char* parameters);
Guolin Ke's avatar
Guolin Ke committed
80

Guolin Ke's avatar
Guolin Ke committed
81
  #pragma region Parameters
82

Guolin Ke's avatar
Guolin Ke committed
83
84
85
  #pragma region Core Parameters

  // [doc-only]
86
87
  // alias = config_file
  // desc = path of config file
88
  // desc = **Note**: can be used only in CLI version
Guolin Ke's avatar
Guolin Ke committed
89
90
91
  std::string config = "";

  // [doc-only]
92
93
94
95
96
97
98
99
  // type = enum
  // default = train
  // options = train, predict, convert_model, refit
  // alias = task_type
  // desc = ``train``, for training, aliases: ``training``
  // desc = ``predict``, for prediction, aliases: ``prediction``, ``test``
  // desc = ``convert_model``, for converting model file into if-else format, see more information in `IO Parameters <#io-parameters>`__
  // desc = ``refit``, for refitting existing models with new data, aliases: ``refit_tree``
Guolin Ke's avatar
Guolin Ke committed
100
  // desc = **Note**: can be used only in CLI version; for language-specific packages you can use the correspondent functions
Guolin Ke's avatar
Guolin Ke committed
101
102
103
  TaskType task = TaskType::kTrain;

  // [doc-only]
104
  // type = enum
105
  // options = regression, regression_l1, huber, fair, poisson, quantile, mape, gamma, tweedie, binary, multiclass, multiclassova, xentropy, xentlambda, lambdarank
106
107
108
109
110
111
112
113
114
115
116
  // alias = objective_type, app, application
  // desc = regression application
  // descl2 = ``regression_l2``, L2 loss, aliases: ``regression``, ``mean_squared_error``, ``mse``, ``l2_root``, ``root_mean_squared_error``, ``rmse``
  // descl2 = ``regression_l1``, L1 loss, aliases: ``mean_absolute_error``, ``mae``
  // descl2 = ``huber``, `Huber loss <https://en.wikipedia.org/wiki/Huber_loss>`__
  // descl2 = ``fair``, `Fair loss <https://www.kaggle.com/c/allstate-claims-severity/discussion/24520>`__
  // descl2 = ``poisson``, `Poisson regression <https://en.wikipedia.org/wiki/Poisson_regression>`__
  // descl2 = ``quantile``, `Quantile regression <https://en.wikipedia.org/wiki/Quantile_regression>`__
  // descl2 = ``mape``, `MAPE loss <https://en.wikipedia.org/wiki/Mean_absolute_percentage_error>`__, aliases: ``mean_absolute_percentage_error``
  // descl2 = ``gamma``, Gamma regression with log-link. It might be useful, e.g., for modeling insurance claims severity, or for any target that might be `gamma-distributed <https://en.wikipedia.org/wiki/Gamma_distribution#Applications>`__
  // descl2 = ``tweedie``, Tweedie regression with log-link. It might be useful, e.g., for modeling total loss in insurance, or for any target that might be `tweedie-distributed <https://en.wikipedia.org/wiki/Tweedie_distribution#Applications>`__
117
  // desc = ``binary``, binary `log loss <https://en.wikipedia.org/wiki/Cross_entropy>`__ classification (or logistic regression). Requires labels in {0, 1}; see ``cross-entropy`` application for general probability labels in [0, 1]
118
119
120
121
122
123
124
125
126
127
128
129
  // desc = multi-class classification application
  // descl2 = ``multiclass``, `softmax <https://en.wikipedia.org/wiki/Softmax_function>`__ objective function, aliases: ``softmax``
  // descl2 = ``multiclassova``, `One-vs-All <https://en.wikipedia.org/wiki/Multiclass_classification#One-vs.-rest>`__ binary objective function, aliases: ``multiclass_ova``, ``ova``, ``ovr``
  // descl2 = ``num_class`` should be set as well
  // desc = cross-entropy application
  // descl2 = ``xentropy``, objective function for cross-entropy (with optional linear weights), aliases: ``cross_entropy``
  // descl2 = ``xentlambda``, alternative parameterization of cross-entropy, aliases: ``cross_entropy_lambda``
  // descl2 = label is anything in interval [0, 1]
  // desc = ``lambdarank``, `lambdarank <https://papers.nips.cc/paper/2971-learning-to-rank-with-nonsmooth-cost-functions.pdf>`__ application
  // descl2 = label should be ``int`` type in lambdarank tasks, and larger number represents the higher relevance (e.g. 0:bad, 1:fair, 2:good, 3:perfect)
  // descl2 = `label_gain <#objective-parameters>`__ can be used to set the gain (weight) of ``int`` label
  // descl2 = all values in ``label`` must be smaller than number of elements in ``label_gain``
Guolin Ke's avatar
Guolin Ke committed
130
131
132
  std::string objective = "regression";

  // [doc-only]
133
134
  // type = enum
  // alias = boosting_type, boost
135
136
137
  // options = gbdt, gbrt, rf, random_forest, dart, goss
  // desc = ``gbdt``, traditional Gradient Boosting Decision Tree, aliases: ``gbrt``
  // desc = ``rf``, Random Forest, aliases: ``random_forest``
138
139
  // desc = ``dart``, `Dropouts meet Multiple Additive Regression Trees <https://arxiv.org/abs/1505.01866>`__
  // desc = ``goss``, Gradient-based One-Side Sampling
Guolin Ke's avatar
Guolin Ke committed
140
141
  std::string boosting = "gbdt";

142
  // alias = train, train_data, train_data_file, data_filename
143
  // desc = path of training data, LightGBM will train from this data
144
  // desc = **Note**: can be used only in CLI version
Guolin Ke's avatar
Guolin Ke committed
145
146
  std::string data = "";

147
  // alias = test, valid_data, valid_data_file, test_data, test_data_file, valid_filenames
148
  // default = ""
149
  // desc = path(s) of validation/test data, LightGBM will output metrics for these data
150
  // desc = support multiple validation data, separated by ``,``
151
  // desc = **Note**: can be used only in CLI version
Guolin Ke's avatar
Guolin Ke committed
152
153
  std::vector<std::string> valid;

154
  // alias = num_iteration, n_iter, num_tree, num_trees, num_round, num_rounds, num_boost_round, n_estimators
155
156
157
  // check = >=0
  // desc = number of boosting iterations
  // desc = **Note**: internally, LightGBM constructs ``num_class * num_iterations`` trees for multi-class classification problems
Guolin Ke's avatar
Guolin Ke committed
158
  int num_iterations = 100;
Guolin Ke's avatar
Guolin Ke committed
159

160
  // alias = shrinkage_rate, eta
161
  // check = >0.0
162
163
  // desc = shrinkage rate
  // desc = in ``dart``, it also affects on normalization weights of dropped trees
Guolin Ke's avatar
Guolin Ke committed
164
165
  double learning_rate = 0.1;

166
  // default = 31
167
  // alias = num_leaf, max_leaves, max_leaf
168
169
  // check = >1
  // desc = max number of leaves in one tree
Guolin Ke's avatar
Guolin Ke committed
170
171
172
  int num_leaves = kDefaultNumLeaves;

  // [doc-only]
173
174
  // type = enum
  // options = serial, feature, data, voting
175
  // alias = tree, tree_type, tree_learner_type
176
177
178
179
180
  // desc = ``serial``, single machine tree learner
  // desc = ``feature``, feature parallel tree learner, aliases: ``feature_parallel``
  // desc = ``data``, data parallel tree learner, aliases: ``data_parallel``
  // desc = ``voting``, voting parallel tree learner, aliases: ``voting_parallel``
  // desc = refer to `Parallel Learning Guide <./Parallel-Learning-Guide.rst>`__ to get more details
Guolin Ke's avatar
Guolin Ke committed
181
182
  std::string tree_learner = "serial";

183
  // alias = num_thread, nthread, nthreads, n_jobs
Guolin Ke's avatar
Guolin Ke committed
184
  // desc = number of threads for LightGBM
185
186
187
188
189
  // desc = ``0`` means default number of threads in OpenMP
  // desc = for the best speed, set this to the number of **real CPU cores**, not the number of threads (most CPUs use `hyper-threading <https://en.wikipedia.org/wiki/Hyper-threading>`__ to generate 2 threads per CPU core)
  // desc = do not set it too large if your dataset is small (for instance, do not use 64 threads for a dataset with 10,000 rows)
  // desc = be aware a task manager or any similar CPU monitoring tool might report that cores not being fully utilized. **This is normal**
  // desc = for parallel learning, do not use all CPU cores because this will cause poor performance for the network communication
Guolin Ke's avatar
Guolin Ke committed
190
191
192
  int num_threads = 0;

  // [doc-only]
193
194
  // type = enum
  // options = cpu, gpu
195
  // alias = device
196
197
198
199
  // desc = device for the tree learning, you can use GPU to achieve the faster learning
  // desc = **Note**: it is recommended to use the smaller ``max_bin`` (e.g. 63) to get the better speed up
  // desc = **Note**: for the faster speed, GPU uses 32-bit float point to sum up by default, so this may affect the accuracy for some tasks. You can set ``gpu_use_dp=true`` to enable 64-bit float point, but it will slow down the training
  // desc = **Note**: refer to `Installation Guide <./Installation-Guide.rst#build-gpu-version>`__ to build LightGBM with GPU support
Guolin Ke's avatar
Guolin Ke committed
200
201
202
  std::string device_type = "cpu";

  // [doc-only]
203
  // alias = random_seed, random_state
204
205
206
207
  // default = None
  // desc = this seed is used to generate other seeds, e.g. ``data_random_seed``, ``feature_fraction_seed``, etc.
  // desc = by default, this seed is unused in favor of default values of other seeds
  // desc = this seed has lower priority in comparison with other seeds, which means that it will be overridden, if you set other seeds explicitly
Guolin Ke's avatar
Guolin Ke committed
208
209
210
211
212
213
  int seed = 0;

  #pragma endregion

  #pragma region Learning Control Parameters

214
  // desc = limit the max depth for tree model. This is used to deal with over-fitting when ``#data`` is small. Tree still grows leaf-wise
215
  // desc = ``<= 0`` means no limit
Guolin Ke's avatar
Guolin Ke committed
216
217
218
  int max_depth = -1;

  // alias = min_data_per_leaf, min_data, min_child_samples
219
220
  // check = >=0
  // desc = minimal number of data in one leaf. Can be used to deal with over-fitting
Guolin Ke's avatar
Guolin Ke committed
221
222
  int min_data_in_leaf = 20;

223
224
225
  // alias = min_sum_hessian_per_leaf, min_sum_hessian, min_hessian, min_child_weight
  // check = >=0.0
  // desc = minimal sum hessian in one leaf. Like ``min_data_in_leaf``, it can be used to deal with over-fitting
Guolin Ke's avatar
Guolin Ke committed
226
227
  double min_sum_hessian_in_leaf = 1e-3;

228
229
230
231
232
233
234
  // alias = sub_row, subsample, bagging
  // check = >0.0
  // check = <=1.0
  // desc = like ``feature_fraction``, but this will randomly select part of data without resampling
  // desc = can be used to speed up training
  // desc = can be used to deal with over-fitting
  // desc = **Note**: to enable bagging, ``bagging_freq`` should be set to a non zero value as well
Guolin Ke's avatar
Guolin Ke committed
235
236
  double bagging_fraction = 1.0;

237
238
239
240
  // alias = subsample_freq
  // desc = frequency for bagging
  // desc = ``0`` means disable bagging; ``k`` means perform bagging at every ``k`` iteration
  // desc = **Note**: to enable bagging, ``bagging_fraction`` should be set to value smaller than ``1.0`` as well
Guolin Ke's avatar
Guolin Ke committed
241
242
243
244
245
246
247
  int bagging_freq = 0;

  // alias = bagging_fraction_seed
  // desc = random seed for bagging
  int bagging_seed = 3;

  // alias = sub_feature, colsample_bytree
248
249
250
251
252
  // check = >0.0
  // check = <=1.0
  // desc = LightGBM will randomly select part of features on each iteration if ``feature_fraction`` smaller than ``1.0``. For example, if you set it to ``0.8``, LightGBM will select 80% of features before training each tree
  // desc = can be used to speed up training
  // desc = can be used to deal with over-fitting
Guolin Ke's avatar
Guolin Ke committed
253
254
  double feature_fraction = 1.0;

255
  // desc = random seed for ``feature_fraction``
Guolin Ke's avatar
Guolin Ke committed
256
257
  int feature_fraction_seed = 2;

258
259
260
  // alias = early_stopping_rounds, early_stopping
  // desc = will stop training if one metric of one validation data doesn't improve in last ``early_stopping_round`` rounds
  // desc = ``<= 0`` means disable
Guolin Ke's avatar
Guolin Ke committed
261
262
  int early_stopping_round = 0;

263
264
265
  // desc = set this to ``true``, if you want to use only the first metric for early stopping
  bool first_metric_only = false;

266
267
268
269
  // alias = max_tree_output, max_leaf_output
  // desc = used to limit the max output of tree leaves
  // desc = ``<= 0`` means no constraint
  // desc = the final max output of leaves is ``learning_rate * max_delta_step``
Guolin Ke's avatar
Guolin Ke committed
270
271
  double max_delta_step = 0.0;

272
273
274
  // alias = reg_alpha
  // check = >=0.0
  // desc = L1 regularization
Guolin Ke's avatar
Guolin Ke committed
275
276
  double lambda_l1 = 0.0;

277
  // alias = reg_lambda, lambda
278
  // check = >=0.0
Guolin Ke's avatar
Guolin Ke committed
279
280
281
  // desc = L2 regularization
  double lambda_l2 = 0.0;

282
283
284
  // alias = min_split_gain
  // check = >=0.0
  // desc = the minimal gain to perform split
Guolin Ke's avatar
Guolin Ke committed
285
286
  double min_gain_to_split = 0.0;

287
  // alias = rate_drop
288
289
290
  // check = >=0.0
  // check = <=1.0
  // desc = used only in ``dart``
291
  // desc = dropout rate: a fraction of previous trees to drop during the dropout
Guolin Ke's avatar
Guolin Ke committed
292
293
  double drop_rate = 0.1;

294
  // desc = used only in ``dart``
295
  // desc = max number of dropped trees during one boosting iteration
296
  // desc = ``<=0`` means no limit
Guolin Ke's avatar
Guolin Ke committed
297
298
  int max_drop = 50;

299
300
301
  // check = >=0.0
  // check = <=1.0
  // desc = used only in ``dart``
302
  // desc = probability of skipping the dropout procedure during a boosting iteration
Guolin Ke's avatar
Guolin Ke committed
303
304
  double skip_drop = 0.5;

305
306
  // desc = used only in ``dart``
  // desc = set this to ``true``, if you want to use xgboost dart mode
Guolin Ke's avatar
Guolin Ke committed
307
308
  bool xgboost_dart_mode = false;

309
310
  // desc = used only in ``dart``
  // desc = set this to ``true``, if you want to use uniform drop
Guolin Ke's avatar
Guolin Ke committed
311
312
  bool uniform_drop = false;

313
314
  // desc = used only in ``dart``
  // desc = random seed to choose dropping models
Guolin Ke's avatar
Guolin Ke committed
315
316
  int drop_seed = 4;

317
318
319
320
  // check = >=0.0
  // check = <=1.0
  // desc = used only in ``goss``
  // desc = the retain ratio of large gradient data
Guolin Ke's avatar
Guolin Ke committed
321
322
  double top_rate = 0.2;

323
324
325
326
  // check = >=0.0
  // check = <=1.0
  // desc = used only in ``goss``
  // desc = the retain ratio of small gradient data
Guolin Ke's avatar
Guolin Ke committed
327
328
  double other_rate = 0.1;

329
330
  // check = >0
  // desc = minimal number of data per categorical group
Guolin Ke's avatar
Guolin Ke committed
331
332
  int min_data_per_group = 100;

333
334
335
  // check = >0
  // desc = used for the categorical features
  // desc = limit the max threshold points in categorical features
Guolin Ke's avatar
Guolin Ke committed
336
337
  int max_cat_threshold = 32;

338
339
340
341
  // check = >=0.0
  // desc = used for the categorical features
  // desc = L2 regularization in categorcial split
  double cat_l2 = 10.0;
Guolin Ke's avatar
Guolin Ke committed
342

343
344
345
346
  // check = >=0.0
  // desc = used for the categorical features
  // desc = this can reduce the effect of noises in categorical features, especially for categories with few data
  double cat_smooth = 10.0;
347

348
349
  // check = >0
  // desc = when number of categories of one feature smaller than or equal to ``max_cat_to_onehot``, one-vs-other split algorithm will be used
Guolin Ke's avatar
Guolin Ke committed
350
351
352
  int max_cat_to_onehot = 4;

  // alias = topk
353
354
355
  // check = >0
  // desc = used in `Voting parallel <./Parallel-Learning-Guide.rst#choose-appropriate-parallel-algorithm>`__
  // desc = set this to larger value for more accurate result, but it will slow down the training speed
Guolin Ke's avatar
Guolin Ke committed
356
357
358
  int top_k = 20;

  // type = multi-int
359
360
361
362
363
  // alias = mc, monotone_constraint
  // default = None
  // desc = used for constraints of monotonic features
  // desc = ``1`` means increasing, ``-1`` means decreasing, ``0`` means non-constraint
  // desc = you need to specify all features in order. For example, ``mc=-1,0,1`` means decreasing for 1st feature, non-constraint for 2nd feature and increasing for the 3rd feature
Guolin Ke's avatar
Guolin Ke committed
364
  std::vector<int8_t> monotone_constraints;
Guolin Ke's avatar
Guolin Ke committed
365
366

  // type = multi-double
367
  // alias = feature_contrib, fc, fp, feature_penalty
Guolin Ke's avatar
Guolin Ke committed
368
369
370
371
  // default = None
  // desc = used to control feature's split gain, will use ``gain[i] = max(0, feature_contri[i]) * gain[i]`` to replace the split gain of i-th feature
  // desc = you need to specify all features in order
  std::vector<double> feature_contri;
372

373
374
375
376
  // alias = fs, forced_splits_filename, forced_splits_file, forced_splits
  // desc = path to a ``.json`` file that specifies splits to force at the top of every decision tree before best-first learning commences
  // desc = ``.json`` file can be arbitrarily nested, and each split contains ``feature``, ``threshold`` fields, as well as ``left`` and ``right`` fields representing subsplits
  // desc = categorical splits are forced in a one-hot fashion, with ``left`` representing the split containing the feature value and ``right`` representing other values
377
  // desc = **Note**: the forced split logic will be ignored, if the split makes gain worse
378
  // desc = see `this file <https://github.com/microsoft/LightGBM/tree/master/examples/binary_classification/forced_splits.json>`__ as an example
Guolin Ke's avatar
Guolin Ke committed
379
380
  std::string forcedsplits_filename = "";

Guolin Ke's avatar
Guolin Ke committed
381
382
383
384
385
386
  // check = >=0.0
  // check = <=1.0
  // desc = decay rate of ``refit`` task, will use ``leaf_output = refit_decay_rate * old_leaf_output + (1.0 - refit_decay_rate) * new_leaf_output`` to refit trees
  // desc = used only in ``refit`` task in CLI version or as argument in ``refit`` function in language-specific package
  double refit_decay_rate = 0.9;

387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
  // check = >=0.0
  // desc = cost-effective gradient boosting multiplier for all penalties
  double cegb_tradeoff = 1.0;

  // check = >=0.0
  // desc = cost-effective gradient-boosting penalty for splitting a node
  double cegb_penalty_split = 0.0;

  // type = multi-double
  // default = 0,0,...,0
  // desc = cost-effective gradient boosting penalty for using a feature
  // desc = applied per data point
  std::vector<double> cegb_penalty_feature_lazy;

  // type = multi-double
  // default = 0,0,...,0
  // desc = cost-effective gradient boosting penalty for using a feature
  // desc = applied once per forest
405
  std::vector<double> cegb_penalty_feature_coupled;
406

Guolin Ke's avatar
Guolin Ke committed
407
408
409
410
  #pragma endregion

  #pragma region IO Parameters

411
412
  // alias = verbose
  // desc = controls the level of LightGBM's verbosity
413
  // desc = ``< 0``: Fatal, ``= 0``: Error (Warning), ``= 1``: Info, ``> 1``: Debug
414
415
416
417
418
419
  int verbosity = 1;

  // check = >1
  // desc = max number of bins that feature values will be bucketed in
  // desc = small number of bins may reduce training accuracy but may increase general power (deal with over-fitting)
  // desc = LightGBM will auto compress memory according to ``max_bin``. For example, LightGBM will use ``uint8_t`` for feature value if ``max_bin=255``
420
  int max_bin = 255;
Guolin Ke's avatar
Guolin Ke committed
421

422
423
424
  // check = >0
  // desc = minimal number of data inside one bin
  // desc = use this to avoid one-data-one-bin (potential over-fitting)
Guolin Ke's avatar
Guolin Ke committed
425
426
  int min_data_in_bin = 3;

427
428
429
430
431
432
433
  // alias = subsample_for_bin
  // check = >0
  // desc = number of data that sampled to construct histogram bins
  // desc = setting this to larger value will give better training result, but will increase data loading time
  // desc = set this to larger value if data is very sparse
  int bin_construct_sample_cnt = 200000;

434
  // alias = hist_pool_size
435
436
437
438
  // desc = max cache size in MB for historical histogram
  // desc = ``< 0`` means no limit
  double histogram_pool_size = -1.0;

439
  // alias = data_seed
440
  // desc = random seed for data partition in parallel learning (excluding the ``feature_parallel`` mode)
Guolin Ke's avatar
Guolin Ke committed
441
  int data_random_seed = 1;
Guolin Ke's avatar
Guolin Ke committed
442

443
444
  // alias = model_output, model_out
  // desc = filename of output model in training
445
  // desc = **Note**: can be used only in CLI version
Guolin Ke's avatar
Guolin Ke committed
446
  std::string output_model = "LightGBM_model.txt";
Guolin Ke's avatar
Guolin Ke committed
447

448
  // alias = save_period
449
450
  // desc = frequency of saving model file snapshot
  // desc = set this to positive value to enable this function. For example, the model file will be snapshotted at each iteration if ``snapshot_freq=1``
451
  // desc = **Note**: can be used only in CLI version
452
453
  int snapshot_freq = -1;

Guolin Ke's avatar
Guolin Ke committed
454
  // alias = model_input, model_in
455
456
457
458
  // desc = filename of input model
  // desc = for ``prediction`` task, this model will be applied to prediction data
  // desc = for ``train`` task, training will be continued from this model
  // desc = **Note**: can be used only in CLI version
Guolin Ke's avatar
Guolin Ke committed
459
  std::string input_model = "";
460

461
  // alias = predict_result, prediction_result, predict_name, prediction_name, pred_name, name_pred
462
  // desc = filename of prediction result in ``prediction`` task
463
  // desc = **Note**: can be used only in CLI version
Guolin Ke's avatar
Guolin Ke committed
464
465
  std::string output_result = "LightGBM_predict_result.txt";

466
  // alias = init_score_filename, init_score_file, init_score, input_init_score
467
  // desc = path of file with training initial scores
468
  // desc = if ``""``, will use ``train_data_file`` + ``.init`` (if exists)
469
  // desc = **Note**: works only in case of loading data directly from file
470
471
472
473
  std::string initscore_filename = "";

  // alias = valid_data_init_scores, valid_init_score_file, valid_init_score
  // default = ""
474
  // desc = path(s) of file(s) with validation initial scores
475
476
  // desc = if ``""``, will use ``valid_data_file`` + ``.init`` (if exists)
  // desc = separate by ``,`` for multi-validation data
477
  // desc = **Note**: works only in case of loading data directly from file
478
479
  std::vector<std::string> valid_data_initscores;

Guolin Ke's avatar
Guolin Ke committed
480
  // alias = is_pre_partition
481
482
  // desc = used for parallel learning (excluding the ``feature_parallel`` mode)
  // desc = ``true`` if training data are pre-partitioned, and different machines use different partitions
Guolin Ke's avatar
Guolin Ke committed
483
484
  bool pre_partition = false;

485
486
487
488
489
490
491
492
493
494
495
496
497
498
  // alias = is_enable_bundle, bundle
  // desc = set this to ``false`` to disable Exclusive Feature Bundling (EFB), which is described in `LightGBM: A Highly Efficient Gradient Boosting Decision Tree <https://papers.nips.cc/paper/6907-lightgbm-a-highly-efficient-gradient-boosting-decision-tree>`__
  // desc = **Note**: disabling this may cause the slow training speed for sparse datasets
  bool enable_bundle = true;

  // check = >=0.0
  // check = <1.0
  // desc = max conflict rate for bundles in EFB
  // desc = set this to ``0.0`` to disallow the conflict and provide more accurate results
  // desc = set this to a larger value to achieve faster speed
  double max_conflict_rate = 0.0;

  // alias = is_sparse, enable_sparse, sparse
  // desc = used to enable/disable sparse optimization
Guolin Ke's avatar
Guolin Ke committed
499
  bool is_enable_sparse = true;
Guolin Ke's avatar
Guolin Ke committed
500

501
502
  // check = >0.0
  // check = <=1.0
503
  // desc = the threshold of zero elements percentage for treating a feature as a sparse one
504
  double sparse_threshold = 0.8;
Guolin Ke's avatar
Guolin Ke committed
505

506
507
508
  // desc = set this to ``false`` to disable the special handle of missing value
  bool use_missing = true;

509
  // desc = set this to ``true`` to treat all zero as missing values (including the unshown values in libsvm/sparse matrices)
510
511
512
513
514
515
  // desc = set this to ``false`` to use ``na`` for representing missing values
  bool zero_as_missing = false;

  // alias = two_round_loading, use_two_round_loading
  // desc = set this to ``true`` if data file is too big to fit in memory
  // desc = by default, LightGBM will map data file to memory and load features from memory. This will provide faster data loading speed, but may cause run out of memory error when the data file is very big
516
  // desc = **Note**: works only in case of loading data directly from file
Guolin Ke's avatar
Guolin Ke committed
517
518
519
  bool two_round = false;

  // alias = is_save_binary, is_save_binary_file
520
  // desc = if ``true``, LightGBM will save the dataset (including validation data) to a binary file. This speed ups the data loading for the next time
521
  // desc = **Note**: can be used only in CLI version; for language-specific packages you can use the correspondent function
Guolin Ke's avatar
Guolin Ke committed
522
523
524
  bool save_binary = false;

  // alias = has_header
525
  // desc = set this to ``true`` if input data has header
526
  // desc = **Note**: works only in case of loading data directly from file
Guolin Ke's avatar
Guolin Ke committed
527
528
  bool header = false;

529
530
531
532
533
  // type = int or string
  // alias = label
  // desc = used to specify the label column
  // desc = use number for index, e.g. ``label=0`` means column\_0 is the label
  // desc = add a prefix ``name:`` for column name, e.g. ``label=name:is_click``
534
  // desc = **Note**: works only in case of loading data directly from file
Guolin Ke's avatar
Guolin Ke committed
535
  std::string label_column = "";
Guolin Ke's avatar
Guolin Ke committed
536

537
538
539
540
541
  // type = int or string
  // alias = weight
  // desc = used to specify the weight column
  // desc = use number for index, e.g. ``weight=0`` means column\_0 is the weight
  // desc = add a prefix ``name:`` for column name, e.g. ``weight=name:weight``
542
  // desc = **Note**: works only in case of loading data directly from file
543
  // desc = **Note**: index starts from ``0`` and it doesn't count the label column when passing type is ``int``, e.g. when label is column\_0, and weight is column\_1, the correct parameter is ``weight=0``
Guolin Ke's avatar
Guolin Ke committed
544
  std::string weight_column = "";
Guolin Ke's avatar
Guolin Ke committed
545

546
547
548
549
550
  // type = int or string
  // alias = group, group_id, query_column, query, query_id
  // desc = used to specify the query/group id column
  // desc = use number for index, e.g. ``query=0`` means column\_0 is the query id
  // desc = add a prefix ``name:`` for column name, e.g. ``query=name:query_id``
551
  // desc = **Note**: works only in case of loading data directly from file
552
553
  // desc = **Note**: data should be grouped by query\_id
  // desc = **Note**: index starts from ``0`` and it doesn't count the label column when passing type is ``int``, e.g. when label is column\_0 and query\_id is column\_1, the correct parameter is ``query=0``
Guolin Ke's avatar
Guolin Ke committed
554
  std::string group_column = "";
Guolin Ke's avatar
Guolin Ke committed
555

556
  // type = multi-int or string
Guolin Ke's avatar
Guolin Ke committed
557
  // alias = ignore_feature, blacklist
558
559
560
561
562
  // desc = used to specify some ignoring columns in training
  // desc = use number for index, e.g. ``ignore_column=0,1,2`` means column\_0, column\_1 and column\_2 will be ignored
  // desc = add a prefix ``name:`` for column name, e.g. ``ignore_column=name:c1,c2,c3`` means c1, c2 and c3 will be ignored
  // desc = **Note**: works only in case of loading data directly from file
  // desc = **Note**: index starts from ``0`` and it doesn't count the label column when passing type is ``int``
563
  // desc = **Note**: despite the fact that specified columns will be completely ignored during the training, they still should have a valid format allowing LightGBM to load file successfully
Guolin Ke's avatar
Guolin Ke committed
564
  std::string ignore_column = "";
565

566
567
568
569
570
571
572
573
  // type = multi-int or string
  // alias = cat_feature, categorical_column, cat_column
  // desc = used to specify categorical features
  // desc = use number for index, e.g. ``categorical_feature=0,1,2`` means column\_0, column\_1 and column\_2 are categorical features
  // desc = add a prefix ``name:`` for column name, e.g. ``categorical_feature=name:c1,c2,c3`` means c1, c2 and c3 are categorical features
  // desc = **Note**: only supports categorical with ``int`` type
  // desc = **Note**: index starts from ``0`` and it doesn't count the label column when passing type is ``int``
  // desc = **Note**: all values should be less than ``Int32.MaxValue`` (2147483647)
574
  // desc = **Note**: using large values could be memory consuming. Tree decision rule works best when categorical features are presented by consecutive integers starting from zero
575
  // desc = **Note**: all negative values will be treated as **missing values**
Guolin Ke's avatar
Guolin Ke committed
576
577
  std::string categorical_feature = "";

578
579
580
581
  // alias = is_predict_raw_score, predict_rawscore, raw_score
  // desc = used only in ``prediction`` task
  // desc = set this to ``true`` to predict only the raw scores
  // desc = set this to ``false`` to predict transformed scores
Guolin Ke's avatar
Guolin Ke committed
582
583
  bool predict_raw_score = false;

584
585
586
  // alias = is_predict_leaf_index, leaf_index
  // desc = used only in ``prediction`` task
  // desc = set this to ``true`` to predict with leaf index of all trees
Guolin Ke's avatar
Guolin Ke committed
587
588
  bool predict_leaf_index = false;

589
590
  // alias = is_predict_contrib, contrib
  // desc = used only in ``prediction`` task
591
  // desc = set this to ``true`` to estimate `SHAP values <https://arxiv.org/abs/1706.06060>`__, which represent how each feature contributes to each prediction
592
  // desc = produces ``#features + 1`` values where the last value is the expected value of the model output over the training data
593
  // desc = **Note**: if you want to get more explanation for your model's predictions using SHAP values like SHAP interaction values, you can install `shap package <https://github.com/slundberg/shap>`__
Nikita Titov's avatar
Nikita Titov committed
594
  // desc = **Note**: unlike the shap package, with ``predict_contrib`` we return a matrix with an extra column, where the last column is the expected value
Guolin Ke's avatar
Guolin Ke committed
595
596
  bool predict_contrib = false;

597
598
599
  // desc = used only in ``prediction`` task
  // desc = used to specify how many trained iterations will be used in prediction
  // desc = ``<= 0`` means no limit
Guolin Ke's avatar
Guolin Ke committed
600
601
  int num_iteration_predict = -1;

602
603
  // desc = used only in ``prediction`` task
  // desc = if ``true``, will use early-stopping to speed up the prediction. May affect the accuracy
604
  bool pred_early_stop = false;
605
606
607

  // desc = used only in ``prediction`` task
  // desc = the frequency of checking early-stopping prediction
608
  int pred_early_stop_freq = 10;
Guolin Ke's avatar
Guolin Ke committed
609

610
611
  // desc = used only in ``prediction`` task
  // desc = the threshold of margin in early-stopping prediction
Guolin Ke's avatar
Guolin Ke committed
612
  double pred_early_stop_margin = 10.0;
Guolin Ke's avatar
Guolin Ke committed
613

614
615
616
  // desc = used only in ``convert_model`` task
  // desc = only ``cpp`` is supported yet
  // desc = if ``convert_model_language`` is set and ``task=train``, the model will be also converted
617
  // desc = **Note**: can be used only in CLI version
Guolin Ke's avatar
Guolin Ke committed
618
619
  std::string convert_model_language = "";

620
621
622
  // alias = convert_model_file
  // desc = used only in ``convert_model`` task
  // desc = output filename of converted model
623
  // desc = **Note**: can be used only in CLI version
Guolin Ke's avatar
Guolin Ke committed
624
625
  std::string convert_model = "gbdt_prediction.cpp";

626
  #pragma endregion
Guolin Ke's avatar
Guolin Ke committed
627
628
629

  #pragma region Objective Parameters

630
631
632
633
  // check = >0
  // alias = num_classes
  // desc = used only in ``multi-class`` classification application
  int num_class = 1;
Guolin Ke's avatar
Guolin Ke committed
634

635
  // alias = unbalance, unbalanced_sets
636
  // desc = used only in ``binary`` application
637
  // desc = set this to ``true`` if training data are unbalanced
638
  // desc = **Note**: while enabling this should increase the overall performance metric of your model, it will also result in poor estimates of the individual class probabilities
639
640
  // desc = **Note**: this parameter cannot be used at the same time with ``scale_pos_weight``, choose only **one** of them
  bool is_unbalance = false;
Guolin Ke's avatar
Guolin Ke committed
641

642
643
644
  // check = >0.0
  // desc = used only in ``binary`` application
  // desc = weight of labels with positive class
645
  // desc = **Note**: while enabling this should increase the overall performance metric of your model, it will also result in poor estimates of the individual class probabilities
646
647
  // desc = **Note**: this parameter cannot be used at the same time with ``is_unbalance``, choose only **one** of them
  double scale_pos_weight = 1.0;
Guolin Ke's avatar
Guolin Ke committed
648

649
650
651
652
  // check = >0.0
  // desc = used only in ``binary`` and ``multiclassova`` classification and in ``lambdarank`` applications
  // desc = parameter for the sigmoid function
  double sigmoid = 1.0;
Guolin Ke's avatar
Guolin Ke committed
653

654
655
  // desc = used only in ``regression``, ``binary`` and ``cross-entropy`` applications
  // desc = adjusts initial score to the mean of labels for faster convergence
Guolin Ke's avatar
Guolin Ke committed
656
657
  bool boost_from_average = true;

658
659
660
661
  // desc = used only in ``regression`` application
  // desc = used to fit ``sqrt(label)`` instead of original values and prediction result will be also automatically converted to ``prediction^2``
  // desc = might be useful in case of large-range labels
  bool reg_sqrt = false;
Guolin Ke's avatar
Guolin Ke committed
662

663
664
665
666
  // check = >0.0
  // desc = used only in ``huber`` and ``quantile`` ``regression`` applications
  // desc = parameter for `Huber loss <https://en.wikipedia.org/wiki/Huber_loss>`__ and `Quantile regression <https://en.wikipedia.org/wiki/Quantile_regression>`__
  double alpha = 0.9;
Guolin Ke's avatar
Guolin Ke committed
667

668
669
670
671
  // check = >0.0
  // desc = used only in ``fair`` ``regression`` application
  // desc = parameter for `Fair loss <https://www.kaggle.com/c/allstate-claims-severity/discussion/24520>`__
  double fair_c = 1.0;
Guolin Ke's avatar
Guolin Ke committed
672

673
674
675
676
  // check = >0.0
  // desc = used only in ``poisson`` ``regression`` application
  // desc = parameter for `Poisson regression <https://en.wikipedia.org/wiki/Poisson_regression>`__ to safeguard optimization
  double poisson_max_delta_step = 0.7;
Guolin Ke's avatar
Guolin Ke committed
677

678
679
680
681
682
683
684
  // check = >=1.0
  // check = <2.0
  // desc = used only in ``tweedie`` ``regression`` application
  // desc = used to control the variance of the tweedie distribution
  // desc = set this closer to ``2`` to shift towards a **Gamma** distribution
  // desc = set this closer to ``1`` to shift towards a **Poisson** distribution
  double tweedie_variance_power = 1.5;
Guolin Ke's avatar
Guolin Ke committed
685

686
687
688
  // check = >0
  // desc = used only in ``lambdarank`` application
  // desc = optimizes `NDCG <https://en.wikipedia.org/wiki/Discounted_cumulative_gain#Normalized_DCG>`__ at this position
Guolin Ke's avatar
Guolin Ke committed
689
  int max_position = 20;
Guolin Ke's avatar
Guolin Ke committed
690

691
692
693
694
695
696
697
  // type = multi-double
  // default = 0,1,3,7,15,31,63,...,2^30-1
  // desc = used only in ``lambdarank`` application
  // desc = relevant gain for labels. For example, the gain of label ``2`` is ``3`` in case of default label gains
  // desc = separate by ``,``
  std::vector<double> label_gain;

Guolin Ke's avatar
Guolin Ke committed
698
699
700
  #pragma endregion

  #pragma region Metric Parameters
701

Guolin Ke's avatar
Guolin Ke committed
702
  // [doc-only]
703
704
705
  // alias = metrics, metric_types
  // default = ""
  // type = multi-enum
706
  // desc = metric(s) to be evaluated on the evaluation set(s)
707
  // descl2 = ``""`` (empty string or not specified) means that metric corresponding to specified ``objective`` will be used (this is possible only for pre-defined objective functions, otherwise no evaluation metric will be added)
708
  // descl2 = ``"None"`` (string, **not** a ``None`` value) means that no metric will be registered, aliases: ``na``, ``null``, ``custom``
709
710
711
712
713
714
715
716
717
718
719
  // descl2 = ``l1``, absolute loss, aliases: ``mean_absolute_error``, ``mae``, ``regression_l1``
  // descl2 = ``l2``, square loss, aliases: ``mean_squared_error``, ``mse``, ``regression_l2``, ``regression``
  // descl2 = ``l2_root``, root square loss, aliases: ``root_mean_squared_error``, ``rmse``
  // descl2 = ``quantile``, `Quantile regression <https://en.wikipedia.org/wiki/Quantile_regression>`__
  // descl2 = ``mape``, `MAPE loss <https://en.wikipedia.org/wiki/Mean_absolute_percentage_error>`__, aliases: ``mean_absolute_percentage_error``
  // descl2 = ``huber``, `Huber loss <https://en.wikipedia.org/wiki/Huber_loss>`__
  // descl2 = ``fair``, `Fair loss <https://www.kaggle.com/c/allstate-claims-severity/discussion/24520>`__
  // descl2 = ``poisson``, negative log-likelihood for `Poisson regression <https://en.wikipedia.org/wiki/Poisson_regression>`__
  // descl2 = ``gamma``, negative log-likelihood for **Gamma** regression
  // descl2 = ``gamma_deviance``, residual deviance for **Gamma** regression
  // descl2 = ``tweedie``, negative log-likelihood for **Tweedie** regression
720
  // descl2 = ``ndcg``, `NDCG <https://en.wikipedia.org/wiki/Discounted_cumulative_gain#Normalized_DCG>`__, aliases: ``lambdarank``
721
722
723
724
725
726
727
728
729
730
  // descl2 = ``map``, `MAP <https://makarandtapaswi.wordpress.com/2012/07/02/intuition-behind-average-precision-and-map/>`__, aliases: ``mean_average_precision``
  // descl2 = ``auc``, `AUC <https://en.wikipedia.org/wiki/Receiver_operating_characteristic#Area_under_the_curve>`__
  // descl2 = ``binary_logloss``, `log loss <https://en.wikipedia.org/wiki/Cross_entropy>`__, aliases: ``binary``
  // descl2 = ``binary_error``, for one sample: ``0`` for correct classification, ``1`` for error classification
  // descl2 = ``multi_logloss``, log loss for multi-class classification, aliases: ``multiclass``, ``softmax``, ``multiclassova``, ``multiclass_ova``, ``ova``, ``ovr``
  // descl2 = ``multi_error``, error rate for multi-class classification
  // descl2 = ``xentropy``, cross-entropy (with optional linear weights), aliases: ``cross_entropy``
  // descl2 = ``xentlambda``, "intensity-weighted" cross-entropy, aliases: ``cross_entropy_lambda``
  // descl2 = ``kldiv``, `Kullback-Leibler divergence <https://en.wikipedia.org/wiki/Kullback%E2%80%93Leibler_divergence>`__, aliases: ``kullback_leibler``
  // desc = support multiple metrics, separated by ``,``
Guolin Ke's avatar
Guolin Ke committed
731
732
  std::vector<std::string> metric;

733
  // check = >0
Guolin Ke's avatar
Guolin Ke committed
734
735
736
737
  // alias = output_freq
  // desc = frequency for metric output
  int metric_freq = 1;

738
739
  // alias = training_metric, is_training_metric, train_metric
  // desc = set this to ``true`` to output metric result over training dataset
740
  // desc = **Note**: can be used only in CLI version
741
  bool is_provide_training_metric = false;
742

743
744
  // type = multi-int
  // default = 1,2,3,4,5
745
  // alias = ndcg_eval_at, ndcg_at, map_eval_at, map_at
746
  // desc = used only with ``ndcg`` and ``map`` metrics
747
  // desc = `NDCG <https://en.wikipedia.org/wiki/Discounted_cumulative_gain#Normalized_DCG>`__ and `MAP <https://makarandtapaswi.wordpress.com/2012/07/02/intuition-behind-average-precision-and-map/>`__ evaluation positions, separated by ``,``
Guolin Ke's avatar
Guolin Ke committed
748
  std::vector<int> eval_at;
Guolin Ke's avatar
Guolin Ke committed
749

Guolin Ke's avatar
Guolin Ke committed
750
751
752
753
  #pragma endregion

  #pragma region Network Parameters

754
755
756
757
  // check = >0
  // alias = num_machine
  // desc = the number of machines for parallel learning application
  // desc = this parameter is needed to be set in both **socket** and **mpi** versions
Guolin Ke's avatar
Guolin Ke committed
758
  int num_machines = 1;
Guolin Ke's avatar
Guolin Ke committed
759

760
761
762
763
  // check = >0
  // alias = local_port, port
  // desc = TCP listen port for local machines
  // desc = **Note**: don't forget to allow this port in firewall settings before training
Guolin Ke's avatar
Guolin Ke committed
764
  int local_listen_port = 12400;
Guolin Ke's avatar
Guolin Ke committed
765

766
767
768
  // check = >0
  // desc = socket time-out in minutes
  int time_out = 120;
Guolin Ke's avatar
Guolin Ke committed
769

770
771
772
  // alias = machine_list_file, machine_list, mlist
  // desc = path of file that lists machines for this parallel learning application
  // desc = each line contains one IP and one port for one machine. The format is ``ip port`` (space as a separator)
Guolin Ke's avatar
Guolin Ke committed
773
  std::string machine_list_filename = "";
Guolin Ke's avatar
Guolin Ke committed
774

775
776
  // alias = workers, nodes
  // desc = list of machines in the following format: ``ip1:port1,ip2:port2``
777
  std::string machines = "";
Guolin Ke's avatar
Guolin Ke committed
778

Guolin Ke's avatar
Guolin Ke committed
779
780
781
782
  #pragma endregion

  #pragma region GPU Parameters

783
784
  // desc = OpenCL platform ID. Usually each GPU vendor exposes one OpenCL platform
  // desc = ``-1`` means the system-wide default platform
785
  // desc = **Note**: refer to `GPU Targets <./GPU-Targets.rst#query-opencl-devices-in-your-system>`__ for more details
Guolin Ke's avatar
Guolin Ke committed
786
787
  int gpu_platform_id = -1;

788
789
  // desc = OpenCL device ID in the specified platform. Each GPU in the selected platform has a unique device ID
  // desc = ``-1`` means the default device in the selected platform
790
  // desc = **Note**: refer to `GPU Targets <./GPU-Targets.rst#query-opencl-devices-in-your-system>`__ for more details
Guolin Ke's avatar
Guolin Ke committed
791
792
  int gpu_device_id = -1;

793
  // desc = set this to ``true`` to use double precision math on GPU (by default single precision is used)
Guolin Ke's avatar
Guolin Ke committed
794
795
796
797
798
  bool gpu_use_dp = false;

  #pragma endregion

  #pragma endregion
Guolin Ke's avatar
Guolin Ke committed
799
800
801

  bool is_parallel = false;
  bool is_parallel_find_bin = false;
Guolin Ke's avatar
Guolin Ke committed
802
803
804
  LIGHTGBM_EXPORT void Set(const std::unordered_map<std::string, std::string>& params);
  static std::unordered_map<std::string, std::string> alias_table;
  static std::unordered_set<std::string> parameter_set;
805

Nikita Titov's avatar
Nikita Titov committed
806
 private:
Guolin Ke's avatar
Guolin Ke committed
807
  void CheckParamConflict();
Guolin Ke's avatar
Guolin Ke committed
808
809
  void GetMembersFromString(const std::unordered_map<std::string, std::string>& params);
  std::string SaveMembersToString() const;
Guolin Ke's avatar
Guolin Ke committed
810
811
};

Guolin Ke's avatar
Guolin Ke committed
812
inline bool Config::GetString(
Guolin Ke's avatar
Guolin Ke committed
813
814
815
816
817
818
819
820
821
  const std::unordered_map<std::string, std::string>& params,
  const std::string& name, std::string* out) {
  if (params.count(name) > 0) {
    *out = params.at(name);
    return true;
  }
  return false;
}

Guolin Ke's avatar
Guolin Ke committed
822
inline bool Config::GetInt(
Guolin Ke's avatar
Guolin Ke committed
823
824
825
  const std::unordered_map<std::string, std::string>& params,
  const std::string& name, int* out) {
  if (params.count(name) > 0) {
826
    if (!Common::AtoiAndCheck(params.at(name).c_str(), out)) {
827
      Log::Fatal("Parameter %s should be of type int, got \"%s\"",
Guolin Ke's avatar
Guolin Ke committed
828
                 name.c_str(), params.at(name).c_str());
829
    }
Guolin Ke's avatar
Guolin Ke committed
830
831
832
833
834
    return true;
  }
  return false;
}

Guolin Ke's avatar
Guolin Ke committed
835
inline bool Config::GetDouble(
Guolin Ke's avatar
Guolin Ke committed
836
  const std::unordered_map<std::string, std::string>& params,
837
  const std::string& name, double* out) {
Guolin Ke's avatar
Guolin Ke committed
838
  if (params.count(name) > 0) {
839
    if (!Common::AtofAndCheck(params.at(name).c_str(), out)) {
840
      Log::Fatal("Parameter %s should be of type double, got \"%s\"",
Guolin Ke's avatar
Guolin Ke committed
841
                 name.c_str(), params.at(name).c_str());
842
    }
Guolin Ke's avatar
Guolin Ke committed
843
844
845
846
847
    return true;
  }
  return false;
}

Guolin Ke's avatar
Guolin Ke committed
848
inline bool Config::GetBool(
Guolin Ke's avatar
Guolin Ke committed
849
850
851
852
  const std::unordered_map<std::string, std::string>& params,
  const std::string& name, bool* out) {
  if (params.count(name) > 0) {
    std::string value = params.at(name);
Guolin Ke's avatar
Guolin Ke committed
853
    std::transform(value.begin(), value.end(), value.begin(), Common::tolower);
854
    if (value == std::string("false") || value == std::string("-")) {
Guolin Ke's avatar
Guolin Ke committed
855
      *out = false;
856
    } else if (value == std::string("true") || value == std::string("+")) {
Guolin Ke's avatar
Guolin Ke committed
857
      *out = true;
858
    } else {
859
      Log::Fatal("Parameter %s should be \"true\"/\"+\" or \"false\"/\"-\", got \"%s\"",
Guolin Ke's avatar
Guolin Ke committed
860
                 name.c_str(), params.at(name).c_str());
Guolin Ke's avatar
Guolin Ke committed
861
862
863
864
865
866
867
868
869
870
    }
    return true;
  }
  return false;
}

struct ParameterAlias {
  static void KeyAliasTransform(std::unordered_map<std::string, std::string>* params) {
    std::unordered_map<std::string, std::string> tmp_map;
    for (const auto& pair : *params) {
Guolin Ke's avatar
Guolin Ke committed
871
      auto alias = Config::alias_table.find(pair.first);
872
      if (alias != Config::alias_table.end()) {  // found alias
Guolin Ke's avatar
Guolin Ke committed
873
        auto alias_set = tmp_map.find(alias->second);
874
875
        if (alias_set != tmp_map.end()) {  // alias already set
                                           // set priority by length & alphabetically to ensure reproducible behavior
wxchan's avatar
wxchan committed
876
877
          if (alias_set->second.size() < pair.first.size() ||
            (alias_set->second.size() == pair.first.size() && alias_set->second < pair.first)) {
878
            Log::Warning("%s is set with %s=%s, %s=%s will be ignored. Current value: %s=%s",
Guolin Ke's avatar
Guolin Ke committed
879
880
                         alias->second.c_str(), alias_set->second.c_str(), params->at(alias_set->second).c_str(),
                         pair.first.c_str(), pair.second.c_str(), alias->second.c_str(), params->at(alias_set->second).c_str());
wxchan's avatar
wxchan committed
881
          } else {
882
            Log::Warning("%s is set with %s=%s, will be overridden by %s=%s. Current value: %s=%s",
Guolin Ke's avatar
Guolin Ke committed
883
884
                         alias->second.c_str(), alias_set->second.c_str(), params->at(alias_set->second).c_str(),
                         pair.first.c_str(), pair.second.c_str(), alias->second.c_str(), pair.second.c_str());
wxchan's avatar
wxchan committed
885
886
            tmp_map[alias->second] = pair.first;
          }
887
        } else {  // alias not set
wxchan's avatar
wxchan committed
888
889
          tmp_map.emplace(alias->second, pair.first);
        }
Guolin Ke's avatar
Guolin Ke committed
890
      } else if (Config::parameter_set.find(pair.first) == Config::parameter_set.end()) {
wxchan's avatar
wxchan committed
891
        Log::Warning("Unknown parameter: %s", pair.first.c_str());
Guolin Ke's avatar
Guolin Ke committed
892
893
894
      }
    }
    for (const auto& pair : tmp_map) {
wxchan's avatar
wxchan committed
895
      auto alias = params->find(pair.first);
896
      if (alias == params->end()) {  // not find
wxchan's avatar
wxchan committed
897
898
899
        params->emplace(pair.first, params->at(pair.second));
        params->erase(pair.second);
      } else {
Guolin Ke's avatar
Guolin Ke committed
900
901
902
        Log::Warning("%s is set=%s, %s=%s will be ignored. Current value: %s=%s",
                     pair.first.c_str(), alias->second.c_str(), pair.second.c_str(), params->at(pair.second).c_str(),
                     pair.first.c_str(), alias->second.c_str());
Guolin Ke's avatar
Guolin Ke committed
903
904
905
906
907
908
909
      }
    }
  }
};

}   // namespace LightGBM

Guolin Ke's avatar
Guolin Ke committed
910
#endif   // LightGBM_CONFIG_H_