RSS metric added (#156)

Author: gyrdym

CHANGELOG.md

@@ -1,5 +1,8 @@
 # Changelog
 
+## 15.3.0
+- RSS metric added
+
 ## 15.2.4
 - Documentation for classification metrics improved
 

lib/src/helpers/validate_matrix_columns.dart

@@ -0,0 +1,13 @@
+import 'package:ml_algo/src/common/exception/matrix_column_exception.dart';
+import 'package:ml_linalg/matrix.dart';
+
+void validateMatrixColumns(Iterable<Matrix> matrices) {
+  final firstInvalidMatrix = matrices
+      .firstWhere((matrix) => matrix.columnsNum != 1, orElse: () => null);
+
+  if (firstInvalidMatrix == null) {
+    return;
+  }
+
+  throw MatrixColumnException(firstInvalidMatrix);
+}

lib/src/metric/metric_type.dart

@@ -1,15 +1,13 @@
 /// Metrics for measuring the quality of the prediction.
 enum MetricType {
-  ///
-  ///
   /// Mean percentage absolute error (MAPE), a regression metric. The formula
   /// is:
   ///
   ///
   /// ![{\mbox{Score}}={\frac{1}{n}}\sum_{{t=1}}^{n}\left|{\frac{y_{t}-\widehat{y}_{t}}{y_{t}}}\right|](https://latex.codecogs.com/gif.latex?%7B%5Cmbox%7BScore%7D%7D%3D%7B%5Cfrac%7B1%7D%7Bn%7D%7D%5Csum_%7B%7Bt%3D1%7D%7D%5E%7Bn%7D%5Cleft%7C%7B%5Cfrac%7By_%7Bt%7D-%5Cwidehat%7By%7D_%7Bt%7D%7D%7By_%7Bt%7D%7D%7D%5Cright%7C)
   ///
   ///
-  /// where y - original value, y with hat - predicted one
+  /// where `y` - original value, `y` with hat - predicted one
   /// 
   ///
   /// The less the score produced by the metric, the better the prediction's
@@ -19,9 +17,7 @@ enum MetricType {
   /// can produce scores which are greater than 1.
   mape,
 
-  ///
-  ///
-  /// Root mean squared error (RMSE), a regression metric. The formula is:
+  /// Root mean squared error (RMSE), a regression metric. The formula is
   ///
   ///
   /// ![{\mbox{Score}}=\sqrt{\frac{1}{n}\sum_{{t=1}}^{n}({\widehat{y}_{t} - y_{t}})^2}](https://latex.codecogs.com/gif.latex?%7B%5Cmbox%7BScore%7D%7D%3D%5Csqrt%7B%5Cfrac%7B1%7D%7Bn%7D%5Csum_%7B%7Bt%3D1%7D%7D%5E%7Bn%7D%28%7B%5Cwidehat%7By%7D_%7Bt%7D%20-%20y_%7Bt%7D%7D%29%5E2%7D)
@@ -35,8 +31,15 @@ enum MetricType {
   /// scores within the range [0, +Infinity]
   rmse,
 
+  /// Residual sum of squares (RSS), a regression metric. The formula is
+  ///
+  /// ![{\mbox{Score}}=\sum_{t=1}^{n}(y_{t} - \widehat{y}_{t})^{2}](https://latex.codecogs.com/gif.latex?%7B%5Cmbox%7BScore%7D%7D%3D%5Csum_%7Bt%3D1%7D%5E%7Bn%7D%28y_%7Bt%7D%20-%20%5Cwidehat%7By%7D_%7Bt%7D%29%5E%7B2%7D)
   ///
+  /// where `n` is a total amount of labels, `y` is an original value, `y` with
+  /// hat - predicted one
   ///
+  rss,
+
   /// A classification metric. The formula is
   ///
   ///
@@ -51,17 +54,15 @@ enum MetricType {
   /// quality is. The metric produces scores within the range [0, 1]
   accuracy,
 
-  ///
-  ///
   /// A classification metric. The formula for a single-class problem is
   ///
   ///
   /// ![{\mbox{Score}}=\frac{TP}{TP + FP}](https://latex.codecogs.com/gif.latex?%7B%5Cmbox%7BScore%7D%7D%3D%5Cfrac%7BTP%7D%7BTP%20+%20FP%7D)
   ///
   ///
-  /// where TP is a number of correctly predicted positive labels (true positive),
-  /// FP - a number of incorrectly predicted positive labels (false positive). In
-  /// other words, TP + FP is a number of all the labels predicted to be positive
+  /// where `TP` is a number of correctly predicted positive labels (true positive),
+  /// `FP` - a number of incorrectly predicted positive labels (false positive). In
+  /// other words, `TP + FP` is a number of all the labels predicted to be positive
   ///
   /// The formula for a multi-class problem is
   ///
@@ -76,17 +77,15 @@ enum MetricType {
   /// range [0, 1]
   precision,
 
-  ///
-  ///
   /// A classification metric. The formula for a single-class problem is
   ///
   ///
   /// ![{\mbox{Score}}=\frac{TP}{TP + FN}](https://latex.codecogs.com/gif.latex?%7B%5Cmbox%7BScore%7D%7D%3D%5Cfrac%7BTP%7D%7BTP%20+%20FN%7D)
   ///
   ///
-  /// where TP is a number of correctly predicted positive labels (true positive),
-  /// FN - a number of incorrectly predicted negative labels (false negative). In
-  /// other words, TP + FN is a total amount of positive labels for a class in
+  /// where `TP` is a number of correctly predicted positive labels (true positive),
+  /// `FN` - a number of incorrectly predicted negative labels (false negative). In
+  /// other words, `TP + FN` is a total amount of positive labels for a class in
   /// the given data
   ///
   /// The formula for a multi-class problem is

lib/src/metric/regression/mape.dart

@@ -1,4 +1,4 @@
-import 'package:ml_algo/src/common/exception/matrix_column_exception.dart';
+import 'package:ml_algo/src/helpers/validate_matrix_columns.dart';
 import 'package:ml_algo/src/metric/metric.dart';
 import 'package:ml_linalg/linalg.dart';
 
@@ -7,16 +7,12 @@ class MapeMetric implements Metric {
 
   @override
   double getScore(Matrix predictedLabels, Matrix originalLabels) {
-    if (predictedLabels.columnsNum != 1) {
-      throw MatrixColumnException(predictedLabels);
-    }
+    validateMatrixColumns([predictedLabels, originalLabels]);
 
-    if (originalLabels.columnsNum != 1) {
-      throw MatrixColumnException(originalLabels);
-    }
-
-    final predicted = predictedLabels.getColumn(0);
-    final original = originalLabels.getColumn(0);
+    final predicted = predictedLabels
+        .toVector();
+    final original = originalLabels
+        .toVector();
 
     return ((original - predicted) / original)
         .abs()

lib/src/metric/regression/rmse.dart

@@ -1,5 +1,6 @@
 import 'dart:math' as math;
 
+import 'package:ml_algo/src/helpers/validate_matrix_columns.dart';
 import 'package:ml_algo/src/metric/metric.dart';
 import 'package:ml_linalg/linalg.dart';
 
@@ -8,16 +9,14 @@ class RmseMetric implements Metric {
 
   @override
   double getScore(Matrix predictedLabels, Matrix origLabels) {
-    if (predictedLabels.columnsNum != 1 || origLabels.columnsNum != 1) {
-      throw Exception('Both predicted labels and original labels have to be '
-          'a matrix-column');
-    }
+    validateMatrixColumns([predictedLabels, origLabels]);
 
     final predicted = predictedLabels
         .getColumn(0);
     final original = origLabels
         .getColumn(0);
 
-    return math.sqrt(((predicted - original).pow(2)).mean());
+    return math
+        .sqrt(((predicted - original).pow(2)).mean());
   }
 }

lib/src/metric/regression/rss.dart

@@ -0,0 +1,21 @@
+import 'package:ml_algo/src/helpers/validate_matrix_columns.dart';
+import 'package:ml_algo/src/metric/metric.dart';
+import 'package:ml_linalg/matrix.dart';
+
+class RssMetric implements Metric {
+  const RssMetric();
+
+  @override
+  double getScore(Matrix predictedLabels, Matrix origLabels) {
+    validateMatrixColumns([predictedLabels, origLabels]);
+
+    final predicted = predictedLabels
+        .toVector();
+    final original = origLabels
+        .toVector();
+
+    return (predicted - original)
+        .pow(2)
+        .sum();
+  }
+}

pubspec.yaml

@@ -1,6 +1,6 @@
 name: ml_algo
 description: Machine learning algorithms, Machine learning models performance evaluation functionality
-version: 15.2.4
+version: 15.3.0
 homepage: https://github.com/gyrdym/ml_algo
 
 environment:

test/metric/regression/rss_test.dart

@@ -0,0 +1,46 @@
+import 'package:ml_algo/src/common/exception/matrix_column_exception.dart';
+import 'package:ml_algo/src/metric/regression/rss.dart';
+import 'package:ml_linalg/matrix.dart';
+import 'package:test/test.dart';
+
+void main() {
+  group('RssMetric', () {
+    const metric = RssMetric();
+    final predictedLabels = Matrix.column([12, 18, 12, 90, 78]);
+    final originalLabels = Matrix.column([10, 20 , 30, 60, 70]);
+
+    test('should throw an error if predicted labels matrix\'s columns count '
+        'is empty', () {
+      final actual = () => metric.getScore(Matrix.empty(), originalLabels);
+
+      expect(actual, throwsA(isA<MatrixColumnException>()));
+    });
+
+    test('should throw an error if predicted labels matrix\'s columns count '
+        'is greater than 1', () {
+      final actual = () => metric.getScore(Matrix.row([1, 2]), originalLabels);
+
+      expect(actual, throwsA(isA<MatrixColumnException>()));
+    });
+
+    test('should throw an error if original labels matrix\'s columns count '
+        'is empty', () {
+      final actual = () => metric.getScore(predictedLabels, Matrix.empty());
+
+      expect(actual, throwsA(isA<MatrixColumnException>()));
+    });
+
+    test('should throw an error if original labels matrix\'s columns count '
+        'is greater than 1', () {
+      final actual = () => metric.getScore(predictedLabels, Matrix.row([1, 2]));
+
+      expect(actual, throwsA(isA<MatrixColumnException>()));
+    });
+
+    test('should count score', () {
+      final actual = metric.getScore(predictedLabels, originalLabels);
+
+      expect(actual, 1296);
+    });
+  });
+}