Erstellen von Histogrammen auf Basis von Statistiken

python/boomer/boosting/cpp/example_wise_statistics.cpp

@@ -273,3 +273,31 @@ AbstractExampleWiseStatistics* DenseExampleWiseStatisticsFactoryImpl::create() {
     return new DenseExampleWiseStatisticsImpl(lossFunctionPtr_, ruleEvaluationPtr_, lapackPtr_, labelMatrixPtr_,
                                               gradients, hessians, currentScores);
 }
+
+DenseExampleWiseStatisticsImpl::DenseExampleWiseStatisticsBinsImpl::DenseExampleWiseStatisticsBinsImpl(DenseExampleWiseStatisticsImpl* statistics, uint32 numBins){
+    statistics_ = statistics;
+    numBins_ = numBins;
+    float64* gradients_ = (float64*)calloc(numBins_, sizeof(float64));
+    float64* hessians_ = (float64*)calloc(numBins_, sizeof(float64));
+    Bin* bins = (Bin*)malloc(numBins_ * sizeof(Bin));
+}
+
+DenseExampleWiseStatisticsImpl::DenseExampleWiseStatisticsBinsImpl::~DenseExampleWiseStatisticsBinsImpl(){
+    free(gradients_);
+    free(hessians_);
+    free(bins);
+}
+
+void DenseExampleWiseStatisticsImpl::DenseExampleWiseStatisticsBinsImpl::onBinUpdate(uint32 binIndex, IndexedFloat32* indexedValue){
+    indexedValue->index = binIndex;
+    bins[binIndex].numExamples++;
+    if(bins[binIndex].maxValue < indexedValue->value){
+        bins[binIndex].maxValue = indexedValue->value;
+    }else if(indexedValue->value < bins[binIndex].minValue){
+        bins[binIndex].minValue = indexedValue->value;
+    }
+}
+
+AbstractStatistics* DenseExampleWiseStatisticsImpl::DenseExampleWiseStatisticsBinsImpl::build(){
+    return statistics_;
+}

python/boomer/boosting/cpp/example_wise_statistics.h

@@ -8,6 +8,7 @@
 
 #include "../../common/cpp/arrays.h"
 #include "../../common/cpp/statistics.h"
+#include "../../common/cpp/binning.h"
 #include "example_wise_rule_evaluation.h"
 #include "example_wise_losses.h"
 #include "statistics.h"
@@ -121,6 +122,32 @@ namespace boosting {
 
             };
 
+            class DenseExampleWiseStatisticsBinsImpl : virtual public IHistogramBuilder {
+
+                private:
+
+                    DenseExampleWiseStatisticsImpl* statistics_;
+
+                    uint32 numBins_;
+
+                    float64* gradients_;
+
+                    float64* hessians_;
+
+                    Bin* bins;
+
+                public:
+
+                    DenseExampleWiseStatisticsBinsImpl(DenseExampleWiseStatisticsImpl* statistics, uint32 numBins);
+
+                    ~DenseExampleWiseStatisticsBinsImpl();
+
+                    void onBinUpdate(uint32 binIndex, IndexedFloat32* indexedValue) override;
+
+                    AbstractStatistics* build();
+
+            };
+
             std::shared_ptr<IExampleWiseLoss> lossFunctionPtr_;
 
             std::shared_ptr<Lapack> lapackPtr_;

python/boomer/boosting/cpp/label_wise_statistics.cpp

@@ -206,3 +206,31 @@ AbstractLabelWiseStatistics* DenseLabelWiseStatisticsFactoryImpl::create() {
     return new DenseLabelWiseStatisticsImpl(lossFunctionPtr_, ruleEvaluationPtr_, labelMatrixPtr_, gradients, hessians,
                                             currentScores);
 }
+
+DenseLabelWiseStatisticsImpl::DenseLabelWiseStatisticsBinsImpl::DenseLabelWiseStatisticsBinsImpl(DenseLabelWiseStatisticsImpl* statistics, uint32 numBins){
+    statistics_ = statistics;
+    numBins_ = numBins;
+    float64* gradients_ = (float64*)calloc(numBins_, sizeof(float64));
+    float64* hessians_ = (float64*)calloc(numBins_, sizeof(float64));
+    Bin* bins = (Bin*)malloc(numBins_ * sizeof(Bin));
+}
+
+DenseLabelWiseStatisticsImpl::DenseLabelWiseStatisticsBinsImpl::~DenseLabelWiseStatisticsBinsImpl(){
+    free(gradients_);
+    free(hessians_);
+    free(bins);
+}
+
+void DenseLabelWiseStatisticsImpl::DenseLabelWiseStatisticsBinsImpl::onBinUpdate(uint32 binIndex, IndexedFloat32* indexedValue){
+    indexedValue->index = binIndex;
+    bins[binIndex].numExamples++;
+    if(bins[binIndex].maxValue < indexedValue->value){
+        bins[binIndex].maxValue = indexedValue->value;
+    }else if(indexedValue->value < bins[binIndex].minValue){
+        bins[binIndex].minValue = indexedValue->value;
+    }
+}
+
+AbstractStatistics* DenseLabelWiseStatisticsImpl::DenseLabelWiseStatisticsBinsImpl::build(){
+    return statistics_;
+}

python/boomer/boosting/cpp/label_wise_statistics.h

@@ -8,6 +8,7 @@
 
 #include "../../common/cpp/arrays.h"
 #include "../../common/cpp/statistics.h"
+#include "../../common/cpp/binning.h"
 #include "label_wise_rule_evaluation.h"
 #include "label_wise_losses.h"
 #include "statistics.h"
@@ -103,6 +104,32 @@ namespace boosting {
 
             };
 
+            class DenseLabelWiseStatisticsBinsImpl : virtual public IHistogramBuilder {
+
+                private:
+
+                    DenseLabelWiseStatisticsImpl* statistics_;
+
+                    uint32 numBins_;
+
+                    float64* gradients_;
+
+                    float64* hessians_;
+
+                    Bin* bins;
+
+                public:
+
+                    DenseLabelWiseStatisticsBinsImpl(DenseLabelWiseStatisticsImpl* statistics, uint32 numBins);
+
+                    ~DenseLabelWiseStatisticsBinsImpl();
+
+                    void onBinUpdate(uint32 binIndex, IndexedFloat32* indexedValue) override;
+
+                    AbstractStatistics* build();
+
+            };
+
             std::shared_ptr<ILabelWiseLoss> lossFunctionPtr_;
 
             std::shared_ptr<IRandomAccessLabelMatrix> labelMatrixPtr_;

python/boomer/common/cpp/binning.h

@@ -51,6 +51,17 @@ class IBinning {
 
 };
 
+/**
+ *  Defines an Interface for methods, which generate histograms, for bins.
+ */
+class IHistogramBuilder : virtual public IBinningObserver {
+
+    public:
+
+        virtual AbstractStatistics* build() = 0;
+
+};
+
 /**
  * Assigns floating point values to bins in a way such that each bin contains approximately the same number of values.
  */

python/boomer/common/cpp/tuples.h

@@ -46,6 +46,14 @@ struct IndexedFloat64 {
     float64 value;
 };
 
+/**
+ *  A struct that stores all necessary information of a group of examples to calculate thresholds.
+ */
+struct Bin {
+    uint32 numExamples;
+    float32 minValue;
+    float32 maxValue;
+};
 
 namespace tuples {