mlpack 4.6.2__cp310-cp310-win_amd64.whl → 4.7.0__cp310-cp310-win_amd64.whl

mlpack/__init__.py

@@ -11,14 +11,14 @@ http://www.opensource.org/licenses/BSD-3-Clause for more information.
 
 
 # start delvewheel patch
-def _delvewheel_patch_1_10_1():
+def _delvewheel_patch_1_12_0():
     import os
     if os.path.isdir(libs_dir := os.path.abspath(os.path.join(os.path.dirname(__file__), os.pardir, 'mlpack.libs'))):
         os.add_dll_directory(libs_dir)
 
 
-_delvewheel_patch_1_10_1()
-del _delvewheel_patch_1_10_1
+_delvewheel_patch_1_12_0()
+del _delvewheel_patch_1_12_0
 # end delvewheel patch
 
 import warnings
@@ -74,4 +74,4 @@ from .adaboost import *
 from .linear_regression_train import linear_regression_train
 from .linear_regression_predict import linear_regression_predict
 from .linear_regression import *
-__version__='4.6.2'
+__version__='4.7.0'

mlpack/include/mlpack/base.hpp

@@ -86,6 +86,7 @@
 #include <armadillo>
 #include <mlpack/core/util/arma_traits.hpp>
 #include <mlpack/core/util/omp_reductions.hpp>
+#include <mlpack/core/arma_extend/find_nan.hpp>
 
 // On Visual Studio, disable C4519 (default arguments for function templates)
 // since it's by default an error, which doesn't even make any sense because

mlpack/include/mlpack/core/arma_extend/find_nan.hpp

@@ -0,0 +1,63 @@
+/**
+ * @file find_nan.hpp
+ * @author Ryan Curtin
+ *
+ * When find_nan() is not available (Armadillo < 11.4), provide an internal
+ * mlpack implementation that operates the same way.  It is slower.
+ */
+#ifndef MLPACK_CORE_ARMA_EXTEND_FIND_NAN_HPP
+#define MLPACK_CORE_ARMA_EXTEND_FIND_NAN_HPP
+
+namespace mlpack {
+
+#if ARMA_VERSION_MAJOR < 11 || \
+    (ARMA_VERSION_MAJOR == 11 && ARMA_VERSION_MINOR < 4)
+
+template<typename T>
+arma::uvec find_nan(const T& m,
+                    const std::enable_if_t<arma::is_arma_type<T>::value>* = 0)
+{
+  typedef typename T::elem_type ElemType;
+
+  if (!std::numeric_limits<ElemType>::has_quiet_NaN)
+    return arma::uvec(); // There can't be any NaNs.
+
+  // find_nonfinite() exists on older Armadillo, and we can also search for +Inf
+  // and -Inf.
+  arma::uvec nonfiniteIndices = arma::find_nonfinite(m);
+  if (nonfiniteIndices.n_elem == 0)
+    return arma::uvec();
+
+  arma::uvec infIndices = arma::find(
+      m == std::numeric_limits<ElemType>::infinity());
+  arma::uvec neginfIndices = arma::find(
+      m == -std::numeric_limits<ElemType>::infinity());
+
+  arma::uvec result(nonfiniteIndices.n_elem -
+      (infIndices.n_elem + neginfIndices.n_elem));
+  if (result.n_elem == 0)
+    return result;
+
+  size_t infIndex = 0;
+  size_t neginfIndex = 0;
+  size_t outputIndex = 0;
+  for (size_t i = 0; i < nonfiniteIndices.n_elem; ++i)
+  {
+    if (infIndex < infIndices.n_elem &&
+        nonfiniteIndices[i] == infIndices[infIndex])
+      ++infIndex;
+    else if (neginfIndex < neginfIndices.n_elem &&
+             nonfiniteIndices[i] == neginfIndices[neginfIndex])
+      ++neginfIndex;
+    else
+      result[outputIndex++] = nonfiniteIndices[i];
+  }
+
+  return result;
+}
+
+#endif
+
+} // namespace mlpack
+
+#endif

mlpack/include/mlpack/core/cereal/low_precision.hpp

@@ -0,0 +1,48 @@
+/**
+ * @file core/cereal/low_precision.hpp
+ * @author Ryan Curtin
+ *
+ * Extra shims necessary for cereal to serialize to JSON for low-precision types
+ * (e.g.  FP16, BF16, etc.).
+ *
+ * mlpack is free software; you may redistribute it and/or modify it under the
+ * terms of the 3-clause BSD license.  You should have received a copy of the
+ * 3-clause BSD license along with mlpack.  If not, see
+ * http://www.opensource.org/licenses/BSD-3-Clause for more information.
+ */
+#ifndef MLPACK_CORE_CEREAL_LOW_PRECISION_HPP
+#define MLPACK_CORE_CEREAL_LOW_PRECISION_HPP
+
+namespace cereal {
+
+// Because our serialization is always done with name-value pairs, we can catch
+// any FP16 serialization at the NVP level with a specialized implementation of
+// the load and save functions for the JSON archive (the only one that does not
+// serialize low-precision correctly).
+
+#if defined(ARMA_HAVE_FP16)
+
+inline void CEREAL_SAVE_FUNCTION_NAME(JSONOutputArchive &ar,
+                                      NameValuePair<arma::fp16&> const& t)
+{
+  ar.setNextName(t.name);
+  std::ostringstream oss;
+  oss.precision(std::numeric_limits<arma::fp16>::max_digits10);
+  oss << t.value;
+  ar(oss.str());
+}
+
+inline void CEREAL_LOAD_FUNCTION_NAME(JSONInputArchive& ar,
+                                      NameValuePair<arma::fp16&>& t)
+{
+  ar.setNextName(t.name);
+  std::string encoded;
+  ar.loadValue(encoded);
+  t.value = arma::fp16(std::stof(encoded));
+}
+
+#endif
+
+} // namespace cereal
+
+#endif

mlpack/include/mlpack/core/cv/cv_base.hpp

@@ -57,12 +57,12 @@ class CVBase
 
   /**
    * Assert that MLAlgorithm takes the numClasses parameter and a
-   * data::DatasetInfo parameter and store them.
+   * DatasetInfo parameter and store them.
    *
    * @param datasetInfo Type information for each dimension of the dataset.
    * @param numClasses Number of classes in the dataset.
    */
-  CVBase(const data::DatasetInfo& datasetInfo,
+  CVBase(const DatasetInfo& datasetInfo,
          const size_t numClasses);
 
   /**
@@ -101,9 +101,9 @@ class CVBase
   static_assert(MIE::IsSupported,
       "The given MLAlgorithm is not supported by MetaInfoExtractor");
 
-  //! A variable for storing a data::DatasetInfo parameter if it is passed.
-  const data::DatasetInfo datasetInfo;
-  //! An indicator whether a data::DatasetInfo parameter has been passed.
+  //! A variable for storing a DatasetInfo parameter if it is passed.
+  const DatasetInfo datasetInfo;
+  //! An indicator whether a DatasetInfo parameter has been passed.
   const bool isDatasetInfoPassed;
   //! A variable for storing the numClasses parameter if it is passed.
   size_t numClasses;
@@ -145,7 +145,7 @@ class CVBase
 
   /**
    * Construct a trained MLAlgorithm model if MLAlgorithm takes the
-   * numClasses parameter and a data::DatasetInfo parameter.
+   * numClasses parameter and a DatasetInfo parameter.
    */
   template<typename... MLAlgorithmArgs,
            bool Enabled = MIE::TakesNumClasses & MIE::TakesDatasetInfo,
@@ -183,7 +183,7 @@ class CVBase
 
   /**
    * Construct a trained MLAlgorithm model if MLAlgorithm takes the
-   * numClasses parameter and a data::DatasetInfo parameter.
+   * numClasses parameter and a DatasetInfo parameter.
    */
   template<typename... MLAlgorithmArgs,
            bool Enabled = MIE::TakesNumClasses & MIE::TakesDatasetInfo,
@@ -196,13 +196,13 @@ class CVBase
                          const MLAlgorithmArgs&... args);
 
   /**
-   * When MLAlgorithm supports a data::DatasetInfo parameter, training should be
+   * When MLAlgorithm supports a DatasetInfo parameter, training should be
    * treated separately - there are models that can be constructed with and
    * without a data:DatasetInfo parameter and models that can be constructed
-   * only with a data::DatasetInfo parameter.
+   * only with a DatasetInfo parameter.
    *
    * Construct a trained MLAlgorithm model when it can be constructed without a
-   * data::DatasetInfo parameter.
+   * DatasetInfo parameter.
    */
   template<bool ConstructableWithoutDatasetInfo,
            typename... MLAlgorithmArgs,
@@ -213,7 +213,7 @@ class CVBase
 
   /**
    * Construct a trained MLAlgorithm model when it can't be constructed without
-   * a data::DatasetInfo parameter.
+   * a DatasetInfo parameter.
    */
   template<bool ConstructableWithoutDatasetInfo,
            typename... MLAlgorithmArgs,

mlpack/include/mlpack/core/cv/cv_base_impl.hpp

@@ -54,7 +54,7 @@ template<typename MLAlgorithm,
 CVBase<MLAlgorithm,
        MatType,
        PredictionsType,
-       WeightsType>::CVBase(const data::DatasetInfo& datasetInfo,
+       WeightsType>::CVBase(const DatasetInfo& datasetInfo,
                             const size_t numClasses) :
     datasetInfo(datasetInfo),
     isDatasetInfoPassed(true),
@@ -63,7 +63,7 @@ CVBase<MLAlgorithm,
   static_assert(MIE::TakesNumClasses,
       "The given MLAlgorithm does not take the numClasses parameter");
   static_assert(MIE::TakesDatasetInfo,
-      "The given MLAlgorithm does not accept a data::DatasetInfo parameter");
+      "The given MLAlgorithm does not accept a DatasetInfo parameter");
 }
 
 template<typename MLAlgorithm,
@@ -184,9 +184,9 @@ MLAlgorithm CVBase<MLAlgorithm,
 {
   static_assert(
       std::is_constructible_v<MLAlgorithm, const MatType&,
-          const data::DatasetInfo, const PredictionsType&, const size_t,
+          const DatasetInfo, const PredictionsType&, const size_t,
               MLAlgorithmArgs...>,
-      "The given MLAlgorithm is not constructible with a data::DatasetInfo "
+      "The given MLAlgorithm is not constructible with a DatasetInfo "
       "parameter and the passed arguments");
 
   static const bool constructableWithoutDatasetInfo =
@@ -256,9 +256,9 @@ MLAlgorithm CVBase<MLAlgorithm,
 {
   static_assert(
       std::is_constructible_v<MLAlgorithm, const MatType&,
-          const data::DatasetInfo, const PredictionsType&, const size_t,
+          const DatasetInfo, const PredictionsType&, const size_t,
               const WeightsType&, MLAlgorithmArgs...>,
-      "The given MLAlgorithm is not constructible with a data::DatasetInfo "
+      "The given MLAlgorithm is not constructible with a DatasetInfo "
       "parameter and the passed arguments");
 
   static const bool constructableWithoutDatasetInfo =
@@ -302,7 +302,7 @@ MLAlgorithm CVBase<MLAlgorithm,
 {
   if (!isDatasetInfoPassed)
     throw std::invalid_argument(
-        "The given MLAlgorithm requires a data::DatasetInfo parameter");
+        "The given MLAlgorithm requires a DatasetInfo parameter");
 
   return MLAlgorithm(xs, datasetInfo, ys, numClasses, args...);
 }

mlpack/include/mlpack/core/cv/k_fold_cv.hpp

@@ -97,7 +97,7 @@ class KFoldCV
 
   /**
    * This constructor can be used for multiclass classification algorithms that
-   * can take a data::DatasetInfo parameter.
+   * can take a DatasetInfo parameter.
    *
    * @param k Number of folds (should be at least 2).
    * @param xs Data points to cross-validate on.
@@ -108,7 +108,7 @@ class KFoldCV
    */
   KFoldCV(const size_t k,
           const MatType& xs,
-          const data::DatasetInfo& datasetInfo,
+          const DatasetInfo& datasetInfo,
           const PredictionsType& ys,
           const size_t numClasses,
           const bool shuffle = true);
@@ -150,7 +150,7 @@ class KFoldCV
 
   /**
    * This constructor can be used for multiclass classification algorithms that
-   * can take a data::DatasetInfo parameter and support weighted learning.
+   * can take a DatasetInfo parameter and support weighted learning.
    *
    * @param k Number of folds (should be at least 2).
    * @param xs Data points to cross-validate on.
@@ -162,7 +162,7 @@ class KFoldCV
    */
   KFoldCV(const size_t k,
           const MatType& xs,
-          const data::DatasetInfo& datasetInfo,
+          const DatasetInfo& datasetInfo,
           const PredictionsType& ys,
           const size_t numClasses,
           const WeightsType& weights,

mlpack/include/mlpack/core/cv/k_fold_cv_impl.hpp

@@ -58,7 +58,7 @@ KFoldCV<MLAlgorithm,
         PredictionsType,
         WeightsType>::KFoldCV(const size_t k,
                               const MatType& xs,
-                              const data::DatasetInfo& datasetInfo,
+                              const DatasetInfo& datasetInfo,
                               const PredictionsType& ys,
                               const size_t numClasses,
                               const bool shuffle) :
@@ -111,7 +111,7 @@ KFoldCV<MLAlgorithm,
         PredictionsType,
         WeightsType>::KFoldCV(const size_t k,
                               const MatType& xs,
-                              const data::DatasetInfo& datasetInfo,
+                              const DatasetInfo& datasetInfo,
                               const PredictionsType& ys,
                               const size_t numClasses,
                               const WeightsType& weights,
@@ -270,7 +270,7 @@ double KFoldCV<MLAlgorithm,
     return 0.0;
   }
 
-  return arma::mean(evaluations.elem(arma::find_finite(evaluations)));
+  return mean(evaluations.elem(find_finite(evaluations)));
 }
 
 template<typename MLAlgorithm,
@@ -300,7 +300,7 @@ double KFoldCV<MLAlgorithm,
       modelPtr.reset(new MLAlgorithm(std::move(model)));
   }
 
-  return arma::mean(evaluations);
+  return mean(evaluations);
 }
 
 template<typename MLAlgorithm,

mlpack/include/mlpack/core/cv/meta_info_extractor.hpp

@@ -25,7 +25,7 @@ namespace mlpack {
  * @tparam MatType The type of data.
  * @tparam PredictionsType The type of predictions.
  * @tparam WeightsType The type of weights.
- * @tparam DatasetInfo An indicator whether a data::DatasetInfo parameter should
+ * @tparam DatasetInfo An indicator whether a DatasetInfo parameter should
  *   be present.
  * @tparam NumClasses An indicator whether the numClasses parameter should be
  *   present.
@@ -101,7 +101,7 @@ struct TrainForm<MT, PT, void, false, false> : public TrainFormBase4<PT, void,
 
 template<typename MT, typename PT>
 struct TrainForm<MT, PT, void, true, false> : public TrainFormBase5<PT, void,
-    const MT&, const data::DatasetInfo&, const PT&> {};
+    const MT&, const DatasetInfo&, const PT&> {};
 
 template<typename MT, typename PT, typename WT>
 struct TrainForm<MT, PT, WT, false, false> : public TrainFormBase5<PT, WT,
@@ -109,7 +109,7 @@ struct TrainForm<MT, PT, WT, false, false> : public TrainFormBase5<PT, WT,
 
 template<typename MT, typename PT, typename WT>
 struct TrainForm<MT, PT, WT, true, false> : public TrainFormBase6<PT, WT,
-    const MT&, const data::DatasetInfo&, const PT&, const WT&> {};
+    const MT&, const DatasetInfo&, const PT&, const WT&> {};
 
 template<typename MT, typename PT>
 struct TrainForm<MT, PT, void, false, true> : public TrainFormBase5<PT, void,
@@ -117,7 +117,7 @@ struct TrainForm<MT, PT, void, false, true> : public TrainFormBase5<PT, void,
 
 template<typename MT, typename PT>
 struct TrainForm<MT, PT, void, true, true> : public TrainFormBase6<PT, void,
-    const MT&, const data::DatasetInfo&, const PT&, const size_t> {};
+    const MT&, const DatasetInfo&, const PT&, const size_t> {};
 
 template<typename MT, typename PT, typename WT>
 struct TrainForm<MT, PT, WT, false, true> : public TrainFormBase6<PT, WT,
@@ -125,7 +125,7 @@ struct TrainForm<MT, PT, WT, false, true> : public TrainFormBase6<PT, WT,
 
 template<typename MT, typename PT, typename WT>
 struct TrainForm<MT, PT, WT, true, true> : public TrainFormBase7<PT, WT,
-    const MT&, const data::DatasetInfo&, const PT&,
+    const MT&, const DatasetInfo&, const PT&,
     const size_t, const WT&> {};
 #else
 template<typename PT, typename WT, typename... SignatureParams>
@@ -147,7 +147,7 @@ struct TrainForm<MT, PT, void, false, false> : public TrainFormBase<PT, void,
 
 template<typename MT, typename PT>
 struct TrainForm<MT, PT, void, true, false> : public TrainFormBase<PT, void,
-    const MT&, const data::DatasetInfo&, const PT&> {};
+    const MT&, const DatasetInfo&, const PT&> {};
 
 template<typename MT, typename PT, typename WT>
 struct TrainForm<MT, PT, WT, false, false> : public TrainFormBase<PT, WT,
@@ -155,7 +155,7 @@ struct TrainForm<MT, PT, WT, false, false> : public TrainFormBase<PT, WT,
 
 template<typename MT, typename PT, typename WT>
 struct TrainForm<MT, PT, WT, true, false> : public TrainFormBase<PT, WT,
-    const MT&, const data::DatasetInfo&, const PT&, const WT&> {};
+    const MT&, const DatasetInfo&, const PT&, const WT&> {};
 
 template<typename MT, typename PT>
 struct TrainForm<MT, PT, void, false, true> : public TrainFormBase<PT, void,
@@ -163,7 +163,7 @@ struct TrainForm<MT, PT, void, false, true> : public TrainFormBase<PT, void,
 
 template<typename MT, typename PT>
 struct TrainForm<MT, PT, void, true, true> : public TrainFormBase<PT, void,
-    const MT&, const data::DatasetInfo&, const PT&, const size_t> {};
+    const MT&, const DatasetInfo&, const PT&, const size_t> {};
 
 template<typename MT, typename PT, typename WT>
 struct TrainForm<MT, PT, WT, false, true> : public TrainFormBase<PT, WT,
@@ -171,7 +171,7 @@ struct TrainForm<MT, PT, WT, false, true> : public TrainFormBase<PT, WT,
 
 template<typename MT, typename PT, typename WT>
 struct TrainForm<MT, PT, WT, true, true> : public TrainFormBase<PT, WT,
-    const MT&, const data::DatasetInfo&, const PT&,
+    const MT&, const DatasetInfo&, const PT&,
     const size_t, const WT&> {};
 #endif
 
@@ -336,7 +336,7 @@ class MetaInfoExtractor
   static const bool SupportsWeights = !std::is_same_v<WeightsType, void*>;
 
   /**
-   * An indication whether MLAlgorithm takes a data::DatasetInfo parameter.
+   * An indication whether MLAlgorithm takes a DatasetInfo parameter.
    */
   static const bool TakesDatasetInfo = Selects<TF5>::value;
 

mlpack/include/mlpack/core/cv/metrics/f1_impl.hpp

@@ -88,7 +88,7 @@ double F1<AS, PC>::Evaluate(MLAlgorithm& model,
         2.0 * precision * recall / (precision + recall);
   }
 
-  return arma::mean(f1s);
+  return mean(f1s);
 }
 
 } // namespace mlpack

mlpack/include/mlpack/core/cv/metrics/facilities.hpp

@@ -28,7 +28,8 @@ template<typename DataType, typename DistanceType>
 DataType PairwiseDistances(const DataType& data,
                            const DistanceType& distance)
 {
-  DataType distances = DataType(data.n_cols, data.n_cols, arma::fill::none);
+  DataType distances = DataType(data.n_cols, data.n_cols,
+      GetFillType<DataType>::none);
   for (size_t i = 0; i < data.n_cols; i++)
   {
     for (size_t j = 0; j < i; j++)

mlpack/include/mlpack/core/cv/metrics/precision_impl.hpp

@@ -77,7 +77,7 @@ double Precision<AS, PC>::Evaluate(MLAlgorithm& model,
     precisions(c) = double(tp) / numberOfPositivePredictions;
   }
 
-  return arma::mean(precisions);
+  return mean(precisions);
 }
 
 } // namespace mlpack

mlpack/include/mlpack/core/cv/metrics/r2_score_impl.hpp

@@ -27,7 +27,7 @@ double R2Score<AdjustedR2>::Evaluate(MLAlgorithm& model,
   // Taking Predicted Output from the model.
   model.Predict(data, predictedResponses);
   // Mean value of response.
-  double meanResponses = arma::mean(responses);
+  double meanResponses = mean(responses);
 
   // Calculate the numerator i.e. residual sum of squares.
   double residualSumSquared = accu(arma::square(responses -

mlpack/include/mlpack/core/cv/metrics/silhouette_score_impl.hpp

@@ -22,7 +22,7 @@ double SilhouetteScore::Overall(const DataType& X,
                                 const Metric& metric)
 {
   util::CheckSameSizes(X, labels, "SilhouetteScore::Overall()");
-  return arma::mean(SamplesScore(X, labels, metric));
+  return mean(SamplesScore(X, labels, metric));
 }
 
 template<typename DataType>

mlpack/include/mlpack/core/cv/simple_cv.hpp

@@ -105,7 +105,7 @@ class SimpleCV
 
   /**
    * This constructor can be used for multiclass classification algorithms that
-   * can take a data::DatasetInfo parameter.
+   * can take a DatasetInfo parameter.
    *
    * @param validationSize A proportion (between 0 and 1) of data used as a
    *     validation set.
@@ -120,7 +120,7 @@ class SimpleCV
   template<typename MatInType, typename PredictionsInType>
   SimpleCV(const double validationSize,
            MatInType&& xs,
-           const data::DatasetInfo& datasetInfo,
+           const DatasetInfo& datasetInfo,
            PredictionsInType&& ys,
            const size_t numClasses);
 
@@ -173,7 +173,7 @@ class SimpleCV
 
   /**
    * This constructor can be used for multiclass classification algorithms that
-   * can take a data::DatasetInfo parameter and support weighted learning.
+   * can take a DatasetInfo parameter and support weighted learning.
    *
    * @param validationSize A proportion (between 0 and 1) of data used as a
    *     validation set.
@@ -192,7 +192,7 @@ class SimpleCV
            typename WeightsInType>
   SimpleCV(const double validationSize,
            MatInType&& xs,
-           const data::DatasetInfo& datasetInfo,
+           const DatasetInfo& datasetInfo,
            PredictionsInType&& ys,
            const size_t numClasses,
            WeightsInType&& weights);

mlpack/include/mlpack/core/cv/simple_cv_impl.hpp

@@ -61,7 +61,7 @@ SimpleCV<MLAlgorithm,
          PredictionsType,
          WeightsType>::SimpleCV(const double validationSize,
                                 MIT&& xs,
-                                const data::DatasetInfo& datasetInfo,
+                                const DatasetInfo& datasetInfo,
                                 PIT&& ys,
                                 const size_t numClasses) :
     SimpleCV(Base(datasetInfo, numClasses), validationSize,
@@ -117,7 +117,7 @@ SimpleCV<MLAlgorithm,
          PredictionsType,
          WeightsType>::SimpleCV(const double validationSize,
                                 MIT&& xs,
-                                const data::DatasetInfo& datasetInfo,
+                                const DatasetInfo& datasetInfo,
                                 PIT&& ys,
                                 const size_t numClasses,
                                 WIT&& weights) :

mlpack/include/mlpack/core/data/binarize.hpp

@@ -16,7 +16,6 @@
 #include <mlpack/prereqs.hpp>
 
 namespace mlpack {
-namespace data {
 
 /**
  * Given an input dataset and threshold, set values greater than threshold to
@@ -86,7 +85,6 @@ void Binarize(const arma::Mat<T>& input,
     output(dimension, i) = input(dimension, i) > threshold;
 }
 
-} // namespace data
 } // namespace mlpack
 
 #endif

mlpack/include/mlpack/core/data/check_categorical_param.hpp

@@ -17,7 +17,6 @@
 #include <mlpack/core/util/params.hpp>
 
 namespace mlpack {
-namespace data {
 
 inline void CheckCategoricalParam(util::Params& params,
                                   const std::string& paramName)
@@ -35,7 +34,6 @@ inline void CheckCategoricalParam(util::Params& params,
     Log::Fatal << errMsg2 << std::endl;
 }
 
-} // namespace data
 } // namespace mlpack
 
 #endif