xgboost
gbm.h
Go to the documentation of this file.
1 
8 #ifndef XGBOOST_GBM_H_
9 #define XGBOOST_GBM_H_
10 
11 #include <dmlc/registry.h>
12 #include <dmlc/any.h>
13 #include <xgboost/base.h>
14 #include <xgboost/data.h>
16 #include <xgboost/model.h>
17 
18 #include <vector>
19 #include <utility>
20 #include <string>
21 #include <functional>
22 #include <unordered_map>
23 #include <memory>
24 
25 namespace xgboost {
26 
27 class Json;
28 class FeatureMap;
29 class ObjFunction;
30 
31 struct GenericParameter;
32 struct LearnerModelParam;
33 struct PredictionCacheEntry;
34 class PredictionContainer;
35 
39 class GradientBooster : public Model, public Configurable {
40  protected:
42  explicit GradientBooster(GenericParameter const* ctx) : ctx_{ctx} {}
43 
44  public:
46  ~GradientBooster() override = default;
53  virtual void Configure(const std::vector<std::pair<std::string, std::string> >& cfg) = 0;
58  virtual void Load(dmlc::Stream* fi) = 0;
63  virtual void Save(dmlc::Stream* fo) const = 0;
71  virtual void Slice(int32_t /*layer_begin*/, int32_t /*layer_end*/, int32_t /*step*/,
72  GradientBooster* /*out*/, bool* /*out_of_bound*/) const {
73  LOG(FATAL) << "Slice is not supported by current booster.";
74  }
77  virtual int32_t BoostedRounds() const = 0;
82  virtual bool ModelFitted() const = 0;
90  virtual void DoBoost(DMatrix* p_fmat, HostDeviceVector<GradientPair>* in_gpair,
91  PredictionCacheEntry*, ObjFunction const* obj) = 0;
92 
102  virtual void PredictBatch(DMatrix* dmat,
103  PredictionCacheEntry* out_preds,
104  bool training,
105  unsigned layer_begin,
106  unsigned layer_end) = 0;
107 
118  virtual void InplacePredict(std::shared_ptr<DMatrix>, float, PredictionCacheEntry*, uint32_t,
119  uint32_t) const {
120  LOG(FATAL) << "Inplace predict is not supported by current booster.";
121  }
134  virtual void PredictInstance(const SparsePage::Inst& inst,
135  std::vector<bst_float>* out_preds,
136  unsigned layer_begin, unsigned layer_end) = 0;
145  virtual void PredictLeaf(DMatrix *dmat,
146  HostDeviceVector<bst_float> *out_preds,
147  unsigned layer_begin, unsigned layer_end) = 0;
148 
160  virtual void PredictContribution(DMatrix* dmat,
161  HostDeviceVector<bst_float>* out_contribs,
162  unsigned layer_begin, unsigned layer_end,
163  bool approximate = false, int condition = 0,
164  unsigned condition_feature = 0) = 0;
165 
167  DMatrix *dmat, HostDeviceVector<bst_float> *out_contribs,
168  unsigned layer_begin, unsigned layer_end, bool approximate) = 0;
169 
177  virtual std::vector<std::string> DumpModel(const FeatureMap& fmap,
178  bool with_stats,
179  std::string format) const = 0;
180 
181  virtual void FeatureScore(std::string const& importance_type,
183  std::vector<bst_feature_t>* features,
184  std::vector<float>* scores) const = 0;
188  virtual bool UseGPU() const = 0;
197  const std::string& name,
198  GenericParameter const* generic_param,
199  LearnerModelParam const* learner_model_param);
200 };
201 
206  : public dmlc::FunctionRegEntryBase<
207  GradientBoosterReg,
208  std::function<GradientBooster*(LearnerModelParam const* learner_model_param,
209  GenericParameter const* ctx)> > {};
210 
223 #define XGBOOST_REGISTER_GBM(UniqueId, Name) \
224  static DMLC_ATTRIBUTE_UNUSED ::xgboost::GradientBoosterReg & \
225  __make_ ## GradientBoosterReg ## _ ## UniqueId ## __ = \
226  ::dmlc::Registry< ::xgboost::GradientBoosterReg>::Get()->__REGISTER__(Name)
227 
228 } // namespace xgboost
229 #endif // XGBOOST_GBM_H_
defines configuration macros of xgboost.
Internal data structured used by XGBoost during training.
Definition: data.h:490
Feature map data structure to help text model dump. TODO(tqchen) consider make it even more lightweig...
Definition: feature_map.h:22
interface of gradient boosting model.
Definition: gbm.h:39
virtual void DoBoost(DMatrix *p_fmat, HostDeviceVector< GradientPair > *in_gpair, PredictionCacheEntry *, ObjFunction const *obj)=0
perform update to the model(boosting)
static GradientBooster * Create(const std::string &name, GenericParameter const *generic_param, LearnerModelParam const *learner_model_param)
create a gradient booster from given name
virtual void Load(dmlc::Stream *fi)=0
load model from stream
~GradientBooster() override=default
virtual destructor
GradientBooster(GenericParameter const *ctx)
Definition: gbm.h:42
virtual void PredictContribution(DMatrix *dmat, HostDeviceVector< bst_float > *out_contribs, unsigned layer_begin, unsigned layer_end, bool approximate=false, int condition=0, unsigned condition_feature=0)=0
feature contributions to individual predictions; the output will be a vector of length (nfeats + 1) *...
virtual void PredictLeaf(DMatrix *dmat, HostDeviceVector< bst_float > *out_preds, unsigned layer_begin, unsigned layer_end)=0
predict the leaf index of each tree, the output will be nsample * ntree vector this is only valid in ...
virtual std::vector< std::string > DumpModel(const FeatureMap &fmap, bool with_stats, std::string format) const =0
dump the model in the requested format
virtual void InplacePredict(std::shared_ptr< DMatrix >, float, PredictionCacheEntry *, uint32_t, uint32_t) const
Inplace prediction.
Definition: gbm.h:118
virtual void FeatureScore(std::string const &importance_type, common::Span< int32_t const > trees, std::vector< bst_feature_t > *features, std::vector< float > *scores) const =0
virtual bool UseGPU() const =0
Whether the current booster uses GPU.
virtual void PredictBatch(DMatrix *dmat, PredictionCacheEntry *out_preds, bool training, unsigned layer_begin, unsigned layer_end)=0
generate predictions for given feature matrix
virtual void Configure(const std::vector< std::pair< std::string, std::string > > &cfg)=0
Set the configuration of gradient boosting. User must call configure once before InitModel and Traini...
virtual bool ModelFitted() const =0
Whether the model has already been trained. When tree booster is chosen, then returns true when there...
virtual void Save(dmlc::Stream *fo) const =0
save model to stream.
virtual void PredictInstance(const SparsePage::Inst &inst, std::vector< bst_float > *out_preds, unsigned layer_begin, unsigned layer_end)=0
online prediction function, predict score for one instance at a time NOTE: use the batch prediction i...
virtual int32_t BoostedRounds() const =0
Return number of boosted rounds.
GenericParameter const * ctx_
Definition: gbm.h:41
virtual void PredictInteractionContributions(DMatrix *dmat, HostDeviceVector< bst_float > *out_contribs, unsigned layer_begin, unsigned layer_end, bool approximate)=0
virtual void Slice(int32_t, int32_t, int32_t, GradientBooster *, bool *) const
Slice a model using boosting index. The slice m:n indicates taking all trees that were fit during the...
Definition: gbm.h:71
Definition: host_device_vector.h:86
interface of objective function
Definition: objective.h:28
span class implementation, based on ISO++20 span<T>. The interface should be the same.
Definition: span.h:423
The input data structure of xgboost.
A device-and-host vector abstraction layer.
Defines the abstract interface for different components in XGBoost.
namespace of xgboost
Definition: base.h:110
Definition: model.h:31
Definition: generic_parameters.h:15
Registry entry for tree updater.
Definition: gbm.h:209
Definition: learner.h:299
Definition: model.h:17
Contains pointer to input matrix and associated cached predictions.
Definition: predictor.h:35