[ML] Allow unbounded num_top_classes in classification analysis (#1526) (#1529)

Author: przemekwitek

include/api/CDataFrameAnalysisConfigReader.h

@@ -84,6 +84,8 @@ class API_EXPORT CDataFrameAnalysisConfigReader {
         bool fallback(bool value) const;
         //! Get an unsigned integer parameter.
         std::size_t fallback(std::size_t value) const;
+        //! Get a signed integer parameter.
+        std::ptrdiff_t fallback(std::ptrdiff_t value) const;
         //! Get a floating point parameter.
         double fallback(double value) const;
         //! Get a string parameter.

include/api/CDataFrameTrainBoostedTreeClassifierRunner.h

@@ -86,7 +86,7 @@ class API_EXPORT CDataFrameTrainBoostedTreeClassifierRunner final
                                      core::CRapidJsonConcurrentLineWriter& writer) const;
 
 private:
-    std::size_t m_NumTopClasses;
+    std::ptrdiff_t m_NumTopClasses;
     EPredictionFieldType m_PredictionFieldType;
     mutable CInferenceModelMetadata m_InferenceModelMetadata;
 };

lib/api/CDataFrameAnalysisConfigReader.cc

@@ -112,6 +112,17 @@ std::size_t CDataFrameAnalysisConfigReader::CParameter::fallback(std::size_t val
     return m_Value->GetUint64();
 }
 
+std::ptrdiff_t CDataFrameAnalysisConfigReader::CParameter::fallback(std::ptrdiff_t value) const {
+    if (m_Value == nullptr) {
+        return value;
+    }
+    if (m_Value->IsInt64() == false) {
+        this->handleFatal();
+        return value;
+    }
+    return m_Value->GetInt64();
+}
+
 double CDataFrameAnalysisConfigReader::CParameter::fallback(double value) const {
     if (m_Value == nullptr) {
         return value;

lib/api/CDataFrameTrainBoostedTreeClassifierRunner.cc

@@ -81,7 +81,7 @@ CDataFrameTrainBoostedTreeClassifierRunner::CDataFrameTrainBoostedTreeClassifier
     : CDataFrameTrainBoostedTreeRunner{
           spec, parameters, loss(parameters[NUM_CLASSES].as<std::size_t>())} {
 
-    m_NumTopClasses = parameters[NUM_TOP_CLASSES].fallback(std::size_t{0});
+    m_NumTopClasses = parameters[NUM_TOP_CLASSES].fallback(std::ptrdiff_t{0});
     m_PredictionFieldType =
         parameters[PREDICTION_FIELD_TYPE].fallback(E_PredictionFieldTypeString);
     this->boostedTreeFactory().classAssignmentObjective(
@@ -138,14 +138,18 @@ void CDataFrameTrainBoostedTreeClassifierRunner::writeOneRow(
     writer.Key(IS_TRAINING_FIELD_NAME);
     writer.Bool(maths::CDataFrameUtils::isMissing(actualClassId) == false);
 
-    if (m_NumTopClasses > 0) {
+    if (m_NumTopClasses != 0) {
         TSizeVec classIds(scores.size());
         std::iota(classIds.begin(), classIds.end(), 0);
         std::sort(classIds.begin(), classIds.end(),
                   [&scores](std::size_t lhs, std::size_t rhs) {
                       return scores[lhs] > scores[rhs];
                   });
-        classIds.resize(std::min(classIds.size(), m_NumTopClasses));
+        // -1 is a special value meaning "output all the classes"
+        classIds.resize(m_NumTopClasses == -1
+                            ? classIds.size()
+                            : std::min(classIds.size(),
+                                       static_cast<std::size_t>(m_NumTopClasses)));
         writer.Key(TOP_CLASSES_FIELD_NAME);
         writer.StartArray();
         for (std::size_t i : classIds) {