Update dimensionality_reduction.py

Author: Sid101098

machine_learning/dimensionality_reduction.py

@@ -179,24 +179,15 @@ def linear_discriminant_analysis(
     """
     assert classes > dimensions
 
-    if features.any():
-        sb = covariance_between_classes(features, labels, classes)
-        sw = covariance_within_classes(features, labels, classes)
-
-        # Add regularization to Sw to avoid singular matrix
-        sw_reg = sw + 1e-6 * np.eye(sw.shape[0])
-
-        # Solve the generalized eigenvalue problem: Sb v = λ Sw v
-        eigenvalues, eigenvectors = eigh(sb, sw_reg)
-
-        # Sort eigenvectors by eigenvalues (descending)
-        idx = np.argsort(eigenvalues)[::-1]
-        eigenvectors = eigenvectors[:, idx]
-
-        # Take top "dimensions" eigenvectors
-        filtered_eigenvectors = eigenvectors[:, :dimensions]
-
-        projected_data = np.dot(filtered_eigenvectors.T, features)
+    if features.any:
+        _, eigenvectors = eigh(
+            covariance_between_classes(features, labels, classes),
+            covariance_within_classes(features, labels, classes),
+        )
+        filtered_eigenvectors = eigenvectors[:, ::-1][:, :dimensions]
+        svd_matrix, _, _ = np.linalg.svd(filtered_eigenvectors)
+        filtered_svd_matrix = svd_matrix[:, 0:dimensions]
+        projected_data = np.dot(filtered_svd_matrix.T, features)
         logging.info("Linear Discriminant Analysis computed")
         return projected_data
     else: