LinearRegressor, default constructor: collectLearningData parameter a…

…dded; dtype parameter passed wherever it's needed

CHANGELOG.md

@@ -1,5 +1,9 @@
 # Changelog
 
+# 13.9.0
+- `LinearRegressor`:
+    - `Default constructor`: `collectLearningData` parameter added
+
 ## 13.8.1
 - `ml_dataframe` dependency updated
 - `xrange` dependency constrain removed

lib/src/classifier/_helpers/log_likelihood_optimizer_factory.dart → lib/src/classifier/_helpers/create_log_likelihood_optimizer.dart

@@ -3,6 +3,8 @@ import 'package:ml_algo/src/cost_function/cost_function_type.dart';
 import 'package:ml_algo/src/di/dependencies.dart';
 import 'package:ml_algo/src/helpers/add_intercept_if.dart';
 import 'package:ml_algo/src/helpers/features_target_split.dart';
+import 'package:ml_algo/src/helpers/normalize_class_labels.dart';
+import 'package:ml_algo/src/helpers/validate_class_labels.dart';
 import 'package:ml_algo/src/linear_optimizer/gradient_optimizer/learning_rate_generator/learning_rate_type.dart';
 import 'package:ml_algo/src/linear_optimizer/initial_coefficients_generator/initial_coefficients_type.dart';
 import 'package:ml_algo/src/linear_optimizer/linear_optimizer.dart';
@@ -33,30 +35,31 @@ LinearOptimizer createLogLikelihoodOptimizer(
   LearningRateType learningRateType,
   InitialCoefficientsType initialWeightsType,
   Matrix initialWeights,
+  num positiveLabel,
+  num negativeLabel,
   DType dtype,
 }) {
-  final splits = featuresTargetSplit(fittingData,
-    targetNames: targetNames,
-  ).toList();
+  validateClassLabels(positiveLabel, negativeLabel);
 
+  final splits = featuresTargetSplit(fittingData, targetNames: targetNames)
+      .toList();
   final points = splits[0].toMatrix(dtype);
   final labels = splits[1].toMatrix(dtype);
-
-  final optimizerFactory = dependencies
-      .getDependency<LinearOptimizerFactory>();
-
-  final costFunctionFactory = dependencies
-      .getDependency<CostFunctionFactory>();
-
+  final optimizerFactory = dependencies.getDependency<LinearOptimizerFactory>();
+  final costFunctionFactory = dependencies.getDependency<CostFunctionFactory>();
   final costFunction = costFunctionFactory.createByType(
     CostFunctionType.logLikelihood,
     linkFunction: linkFunction,
+    positiveLabel: positiveLabel,
+    negativeLabel: negativeLabel,
   );
+  final normalizedLabels = normalizeClassLabels(labels,
+      positiveLabel, negativeLabel);
 
   return optimizerFactory.createByType(
     optimizerType,
     addInterceptIf(fitIntercept, points, interceptScale, dtype),
-    labels,
+    normalizedLabels,
     costFunction: costFunction,
     iterationLimit: iterationsLimit,
     initialLearningRate: initialLearningRate,

lib/src/classifier/_mixins/linear_classifier_mixin.dart

@@ -23,6 +23,7 @@ mixin LinearClassifierMixin implements LinearClassifier {
       fitIntercept,
       testFeatures.toMatrix(dtype),
       interceptScale,
+      dtype,
     );
 
     validateCoefficientsMatrix(coefficientsByClasses,

lib/src/classifier/logistic_regressor/_helpers/create_logistic_regressor.dart

@@ -1,9 +1,9 @@
-import 'package:ml_algo/src/classifier/_helpers/log_likelihood_optimizer_factory.dart';
+import 'package:ml_algo/src/classifier/_helpers/create_log_likelihood_optimizer.dart';
 import 'package:ml_algo/src/classifier/logistic_regressor/logistic_regressor.dart';
-import 'package:ml_algo/src/classifier/logistic_regressor/logistic_regressor_factory.dart';
+import 'package:ml_algo/src/classifier/logistic_regressor/logistic_regressor_impl.dart';
 import 'package:ml_algo/src/di/dependencies.dart';
+import 'package:ml_algo/src/helpers/validate_class_labels.dart';
 import 'package:ml_algo/src/helpers/validate_initial_coefficients.dart';
-import 'package:ml_algo/src/helpers/validate_probability_threshold.dart';
 import 'package:ml_algo/src/helpers/validate_train_data.dart';
 import 'package:ml_algo/src/linear_optimizer/gradient_optimizer/learning_rate_generator/learning_rate_type.dart';
 import 'package:ml_algo/src/linear_optimizer/initial_coefficients_generator/initial_coefficients_type.dart';
@@ -36,10 +36,11 @@ LogisticRegressor createLogisticRegressor(
     Vector initialCoefficients,
     num positiveLabel,
     num negativeLabel,
+    bool collectLearningData,
     DType dtype,
 ) {
-  validateProbabilityThreshold(probabilityThreshold);
   validateTrainData(trainData, [targetName]);
+  validateClassLabels(positiveLabel, negativeLabel);
 
   if (initialCoefficients.isNotEmpty) {
     validateInitialCoefficients(initialCoefficients, fitIntercept,
@@ -48,7 +49,6 @@ LogisticRegressor createLogisticRegressor(
 
   final linkFunction = dependencies.getDependency<LinkFunction>(
       dependencyName: dTypeToInverseLogitLinkFunctionToken[dtype]);
-
   final optimizer = createLogLikelihoodOptimizer(
     trainData,
     [targetName],
@@ -66,28 +66,31 @@ LogisticRegressor createLogisticRegressor(
     fitIntercept: fitIntercept,
     interceptScale: interceptScale,
     isFittingDataNormalized: isFittingDataNormalized,
+    positiveLabel: positiveLabel,
+    negativeLabel: negativeLabel,
     dtype: dtype,
   );
-
   final coefficientsByClasses = optimizer.findExtrema(
     initialCoefficients: initialCoefficients.isNotEmpty
         ? Matrix.fromColumns([initialCoefficients], dtype: dtype)
         : null,
     isMinimizingObjective: false,
+    collectLearningData: collectLearningData,
   );
+  final costPerIteration = optimizer.costPerIteration.isNotEmpty
+      ? optimizer.costPerIteration
+      : null;
 
-  final regressorFactory = dependencies
-      .getDependency<LogisticRegressorFactory>();
-
-  return regressorFactory.create(
-    targetName,
+  return LogisticRegressorImpl(
+    [targetName],
     linkFunction,
-    probabilityThreshold,
     fitIntercept,
     interceptScale,
     coefficientsByClasses,
+    probabilityThreshold,
     negativeLabel,
     positiveLabel,
+    costPerIteration,
     dtype,
   );
 }

lib/src/classifier/logistic_regressor/_helpers/create_logistic_regressor_from_json.dart

@@ -5,5 +5,6 @@ import 'package:ml_algo/src/classifier/logistic_regressor/logistic_regressor_imp
 
 LogisticRegressor createLogisticRegressorFromJson(String json) {
   final decoded = jsonDecode(json) as Map<String, dynamic>;
+
   return LogisticRegressorImpl.fromJson(decoded);
 }

lib/src/classifier/logistic_regressor/logistic_regressor.dart

@@ -13,10 +13,10 @@ import 'package:ml_linalg/vector.dart';
 
 /// Logistic regression-based classification.
 ///
-/// Logistic regression is an algorithm that solves a binary classification
+/// Logistic regression is an algorithm that solves the binary classification
 /// problem. The algorithm uses maximization of the passed data likelihood.
 /// In other words, the regressor iteratively tries to select coefficients
-/// that makes combination of passed features and their coefficients most
+/// that makes combination of passed features and these coefficients most
 /// likely.
 abstract class LogisticRegressor implements
     LinearClassifier, Assessable, Serializable {
@@ -112,6 +112,11 @@ abstract class LogisticRegressor 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]
@@ -136,6 +141,7 @@ abstract class LogisticRegressor implements
     Vector initialCoefficients,
     num positiveLabel = 1,
     num negativeLabel = 0,
+    bool collectLearningData = false,
     DType dtype = DType.float32,
   }) => createLogisticRegressor(
     trainData,
@@ -157,6 +163,7 @@ abstract class LogisticRegressor implements
     initialCoefficients ?? Vector.empty(dtype: dtype),
     positiveLabel,
     negativeLabel,
+    collectLearningData,
     dtype,
   );
 
@@ -200,4 +207,8 @@ abstract class LogisticRegressor implements
   /// ````
   factory LogisticRegressor.fromJson(String json) =>
       createLogisticRegressorFromJson(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/logistic_regressor/logistic_regressor_impl.dart

@@ -3,7 +3,9 @@ import 'package:ml_algo/src/classifier/_mixins/linear_classifier_mixin.dart';
 import 'package:ml_algo/src/classifier/logistic_regressor/logistic_regressor.dart';
 import 'package:ml_algo/src/classifier/logistic_regressor/logistic_regressor_json_keys.dart';
 import 'package:ml_algo/src/common/serializable/serializable_mixin.dart';
+import 'package:ml_algo/src/helpers/validate_class_labels.dart';
 import 'package:ml_algo/src/helpers/validate_coefficients_matrix.dart';
+import 'package:ml_algo/src/helpers/validate_probability_threshold.dart';
 import 'package:ml_algo/src/link_function/helpers/from_link_function_json.dart';
 import 'package:ml_algo/src/link_function/helpers/link_function_to_json.dart';
 import 'package:ml_algo/src/link_function/link_function.dart';
@@ -15,7 +17,6 @@ import 'package:ml_linalg/from_dtype_json.dart';
 import 'package:ml_linalg/from_matrix_json.dart';
 import 'package:ml_linalg/matrix.dart';
 import 'package:ml_linalg/matrix_to_json.dart';
-import 'package:ml_linalg/vector.dart';
 
 part 'logistic_regressor_impl.g.dart';
 
@@ -37,8 +38,11 @@ class LogisticRegressorImpl
       this.probabilityThreshold,
       this.negativeLabel,
       this.positiveLabel,
+      this.costPerIteration,
       this.dtype,
   ) {
+    validateProbabilityThreshold(probabilityThreshold);
+    validateClassLabels(positiveLabel, negativeLabel);
     validateCoefficientsMatrix(coefficientsByClasses);
 
     // Logistic regression specific check, it cannot be placed in
@@ -105,23 +109,25 @@ class LogisticRegressorImpl
   final LinkFunction linkFunction;
 
   @override
-  DataFrame predict(DataFrame testFeatures) {
-    final probabilities = getProbabilitiesMatrix(testFeatures).getColumn(0);
-
-    final classesList = probabilities
-        // TODO: use SIMD
-        .map((value) => value >= probabilityThreshold
-          ? positiveLabel
-          : negativeLabel,
-        )
-        .toList(growable: false);
+  @JsonKey(
+    name: logisticRegressorCostPerIterationJsonKey,
+    includeIfNull: false,
+  )
+  final List<num> costPerIteration;
 
-    final classesMatrix = Matrix.fromColumns([
-      Vector.fromList(classesList),
-    ]);
+  @override
+  DataFrame predict(DataFrame testFeatures) {
+    final predictedLabels = getProbabilitiesMatrix(testFeatures)
+        .mapColumns(
+            (column) => column.mapToVector(
+                    (probability) => probability >= probabilityThreshold
+                        ? positiveLabel.toDouble()
+                        : negativeLabel.toDouble()
+            ),
+    );
 
     return DataFrame.fromMatrix(
-      classesMatrix,
+      predictedLabels,
       header: classNames,
     );
   }

lib/src/classifier/logistic_regressor/logistic_regressor_json_keys.dart

@@ -7,3 +7,4 @@ const logisticRegressorProbabilityThresholdJsonKey = 'PT';
 const logisticRegressorPositiveLabelJsonKey = 'PL';
 const logisticRegressorNegativeLabelJsonKey = 'NL';
 const logisticRegressorLinkFunctionJsonKey = 'LF';
+const logisticRegressorCostPerIterationJsonKey = 'CPI';