collectLearningData argument added to SoftmaxRegressor default constr…

…uctor

CHANGELOG.md

@@ -1,5 +1,9 @@
 # Changelog
 
+## 14.2.0
+- `SoftmaxRegressor`:
+    - `Default constructor`: `collectLearningData` parameter added
+
 ## 14.1.1
 - `README`: Advanced usage example on Logistic regression added
 

lib/src/classifier/softmax_regressor/_helpers/create_softmax_regressor.dart

@@ -32,6 +32,7 @@ SoftmaxRegressor createSoftmaxRegressor(
     Matrix initialCoefficients,
     num positiveLabel,
     num negativeLabel,
+    bool collectLearningData,
     DType dtype,
 ) {
   if (targetNames.isNotEmpty && targetNames.length < 2) {
@@ -69,7 +70,9 @@ SoftmaxRegressor createSoftmaxRegressor(
   final coefficientsByClasses = optimizer.findExtrema(
     initialCoefficients: initialCoefficients,
     isMinimizingObjective: false,
+    collectLearningData: collectLearningData,
   );
+  final costPerIteration = optimizer.costPerIteration;
 
   final regressorFactory = dependencies
       .getDependency<SoftmaxRegressorFactory>();
@@ -82,6 +85,7 @@ SoftmaxRegressor createSoftmaxRegressor(
     interceptScale,
     positiveLabel,
     negativeLabel,
+    costPerIteration,
     dtype,
   );
 }

lib/src/classifier/softmax_regressor/softmax_regressor.dart

@@ -105,6 +105,11 @@ abstract class SoftmaxRegressor implements
   /// [negativeLabel] Defines the value that will be used for `negative` class.
   /// By default, `0`.
   ///
+  /// [collectLearningData] Whether or not to collect learning data, for
+  /// instance cost function value per each iteration. Affects performance much.
+  /// If [collectLearningData] is true, one may access [costPerIteration]
+  /// getter in order to evaluate learning process more thoroughly.
+  ///
   /// [dtype] A data type for all the numeric values, used by the algorithm. Can
   /// affect performance or accuracy of the computations. Default value is
   /// [DType.float32]
@@ -129,6 +134,7 @@ abstract class SoftmaxRegressor implements
         Matrix initialCoefficients,
         num positiveLabel = 1,
         num negativeLabel = 0,
+        bool collectLearningData = false,
         DType dtype = DType.float32,
       }
   ) => createSoftmaxRegressor(
@@ -150,6 +156,7 @@ abstract class SoftmaxRegressor implements
     initialCoefficients,
     positiveLabel,
     negativeLabel,
+    collectLearningData,
     dtype,
   );
 
@@ -192,4 +199,8 @@ abstract class SoftmaxRegressor implements
   /// ````
   factory SoftmaxRegressor.fromJson(String json) =>
       createSoftmaxRegressorFromJson(json);
+
+  /// Returns a list of cost values per each learning iteration. Returns null
+  /// if the parameter `collectLearningData` of the default constructor is false
+  List<num> get costPerIteration;
 }

lib/src/classifier/softmax_regressor/softmax_regressor_factory.dart

@@ -12,6 +12,7 @@ abstract class SoftmaxRegressorFactory {
       num interceptScale,
       num positiveLabel,
       num negativeLabel,
+      List<num> costPerIteration,
       DType dtype,
   );
 }

lib/src/classifier/softmax_regressor/softmax_regressor_factory_impl.dart

@@ -17,6 +17,7 @@ class SoftmaxRegressorFactoryImpl implements SoftmaxRegressorFactory {
       num interceptScale,
       num positiveLabel,
       num negativeLabel,
+      List<num> costPerIteration,
       DType dtype,
   ) => SoftmaxRegressorImpl(
     coefficientsByClasses,
@@ -26,6 +27,7 @@ class SoftmaxRegressorFactoryImpl implements SoftmaxRegressorFactory {
     interceptScale,
     positiveLabel,
     negativeLabel,
+    costPerIteration,
     dtype,
   );
 }

lib/src/classifier/softmax_regressor/softmax_regressor_impl.dart

@@ -33,6 +33,7 @@ class SoftmaxRegressorImpl
       this.interceptScale,
       this.positiveLabel,
       this.negativeLabel,
+      this.costPerIteration,
       this.dtype,
   ) {
     validateClassLabels(positiveLabel, negativeLabel);
@@ -94,6 +95,13 @@ class SoftmaxRegressorImpl
   @JsonKey(name: softmaxRegressorNegativeLabelJsonKey)
   final num negativeLabel;
 
+  @override
+  @JsonKey(
+    name: softmaxRegressorCostPerIterationJsonKey,
+    includeIfNull: false,
+  )
+  final List<num> costPerIteration;
+
   @override
   DataFrame predict(DataFrame testFeatures) {
     final allProbabilities = getProbabilitiesMatrix(testFeatures);

lib/src/classifier/softmax_regressor/softmax_regressor_json_keys.dart

@@ -6,3 +6,4 @@ const softmaxRegressorDTypeJsonKey = 'DT';
 const softmaxRegressorLinkFunctionJsonKey = 'LF';
 const softmaxRegressorPositiveLabelJsonKey = 'PL';
 const softmaxRegressorNegativeLabelJsonKey = 'NL';
+const softmaxRegressorCostPerIterationJsonKey = 'CPI';

pubspec.yaml

@@ -1,6 +1,6 @@
 name: ml_algo
 description: Machine learning algorithms, Machine learning models performance evaluation functionality
-version: 14.1.1
+version: 14.2.0
 homepage: https://github.com/gyrdym/ml_algo
 
 environment:

test/classifier/softmax_regressor/unit_test/softmax_regressor_factory_impl_test.dart

@@ -22,6 +22,7 @@ void main() {
       final interceptScale = 1;
       final positiveLabel = 1;
       final negativeLabel = -1;
+      final costPerIteration = [1, 2, 3];
       final dtype = DType.float32;
 
       final regressor = factory.create(
@@ -32,6 +33,7 @@ void main() {
         interceptScale,
         positiveLabel,
         negativeLabel,
+        costPerIteration,
         dtype,
       );
 

test/classifier/softmax_regressor/unit_test/softmax_regressor_impl_test.dart

@@ -16,6 +16,7 @@ void main() {
     final interceptScale = 10;
     final positiveLabel = 1.0;
     final negativeLabel = -1.0;
+    final costPerIteration = [10, -10, 20, 2.3];
     final dtype = DType.float32;
 
     final coefficientsByClasses = Matrix.fromList([
@@ -91,6 +92,7 @@ void main() {
         interceptScale,
         positiveLabel,
         negativeLabel,
+        costPerIteration,
         dtype,
       );
     });
@@ -109,6 +111,7 @@ void main() {
           interceptScale,
           positiveLabel,
           negativeLabel,
+          costPerIteration,
           dtype,
         );
 
@@ -129,6 +132,7 @@ void main() {
           interceptScale,
           positiveLabel,
           negativeLabel,
+          costPerIteration,
           dtype,
         );
 

test/mocks.dart

@@ -252,7 +252,7 @@ KnnRegressorFactory createKnnRegressorFactoryMock(KnnRegressor regressor) {
 SoftmaxRegressorFactory createSoftmaxRegressorFactoryMock(
     SoftmaxRegressor softmaxRegressor) {
   final factory = SoftmaxRegressorFactoryMock();
-  when(factory.create(any, any, any, any, any, any, any, any))
+  when(factory.create(any, any, any, any, any, any, any, any, any))
       .thenReturn(softmaxRegressor);
   return factory;
 }