fix: accuracy_score not computing properly for binary classification

neuralnetlib/metrics.py

@@ -1,27 +1,29 @@
 import numpy as np
+
 from neuralnetlib.utils import apply_threshold
 
-def accuracy_score(y_pred: np.ndarray, y_true: np.ndarray) -> float:
-    if y_pred.shape[1] == 1:  # Binary classification
-        y_pred_classes = apply_threshold(y_pred)
+
+def accuracy_score(y_pred: np.ndarray, y_true: np.ndarray, threshold: float = 0.5) -> float:
+    if y_pred.ndim == 1 or y_pred.shape[1] == 1:  # Binary classification
+        y_pred_classes = apply_threshold(y_pred, threshold).ravel()
     else:  # Multiclass classification
         y_pred_classes = np.argmax(y_pred, axis=1)
+
     if y_true.ndim == 1 or y_true.shape[1] == 1:  # If y_true is not one-hot encoded
         y_true_classes = y_true.ravel()
     else:
         y_true_classes = np.argmax(y_true, axis=1)
-
-    return np.mean(y_pred_classes == y_true_classes)
 
+    return np.mean(y_pred_classes == y_true_classes)
 
-def f1_score(y_pred: np.ndarray, y_true: np.ndarray) -> float:
-    precision = precision_score(y_pred, y_true)
-    recall = recall_score(y_pred, y_true)
+def f1_score(y_pred: np.ndarray, y_true: np.ndarray, threshold: float = 0.5) -> float:
+    precision = precision_score(y_pred, y_true, threshold)
+    recall = recall_score(y_pred, y_true, threshold)
     return 2 * (precision * recall) / (precision + recall) if (precision + recall) > 0 else 0
 
 
-def recall_score(y_pred: np.ndarray, y_true: np.ndarray) -> float:
-    y_pred_labels = (y_pred >= 0.5).astype(int) if y_pred.shape[1] == 1 else np.argmax(y_pred, axis=1)
+def recall_score(y_pred: np.ndarray, y_true: np.ndarray, threshold: float = 0.5) -> float:
+    y_pred_labels = apply_threshold(y_pred, threshold) if y_pred.shape[1] == 1 else np.argmax(y_pred, axis=1)
     y_true_labels = y_true if y_true.ndim == 1 or y_true.shape[1] == 1 else np.argmax(y_true, axis=1)
     classes = np.unique(y_true_labels)
     recall_scores = []
@@ -35,8 +37,9 @@ def recall_score(y_pred: np.ndarray, y_true: np.ndarray) -> float:
 
     return np.mean(recall_scores)
 
-def precision_score(y_pred: np.ndarray, y_true: np.ndarray) -> float:
-    y_pred_labels = (y_pred >= 0.5).astype(int) if y_pred.shape[1] == 1 else np.argmax(y_pred, axis=1)
+
+def precision_score(y_pred: np.ndarray, y_true: np.ndarray, threshold: float = 0.5) -> float:
+    y_pred_labels = apply_threshold(y_pred, threshold) if y_pred.shape[1] == 1 else np.argmax(y_pred, axis=1)
     y_true_labels = y_true if y_true.ndim == 1 or y_true.shape[1] == 1 else np.argmax(y_true, axis=1)
     classes = np.unique(y_true_labels)
     precision_scores = []