Upload src/evaluation/metrics.py with huggingface_hub

src/evaluation/metrics.py

@@ -0,0 +1,86 @@
+"""
+Evaluation metrics for anomaly detection performance.
+
+Computes accuracy, precision, recall, F1-score, and AUC-ROC.
+"""
+
+import numpy as np
+from typing import Optional, List
+from sklearn.metrics import (
+    accuracy_score,
+    precision_score,
+    recall_score,
+    f1_score,
+    roc_auc_score,
+    confusion_matrix,
+    classification_report,
+)
+
+
+class MetricsCalculator:
+    """Calculates all evaluation metrics for binary classification."""
+
+    @staticmethod
+    def compute_all(
+        y_true: list | np.ndarray,
+        y_pred: list | np.ndarray,
+        y_scores: Optional[list | np.ndarray] = None,
+    ) -> dict:
+        """
+        Compute all metrics.
+
+        Args:
+            y_true:   Ground-truth labels (0 = normal, 1 = abnormal).
+            y_pred:   Predicted labels.
+            y_scores: Predicted probabilities for the positive class
+                      (required for AUC-ROC).
+
+        Returns:
+            Dictionary of metric_name → value.
+        """
+        y_true = np.asarray(y_true)
+        y_pred = np.asarray(y_pred)
+
+        metrics = {
+            "accuracy": accuracy_score(y_true, y_pred),
+            "precision": precision_score(y_true, y_pred, zero_division=0),
+            "recall": recall_score(y_true, y_pred, zero_division=0),
+            "f1_score": f1_score(y_true, y_pred, zero_division=0),
+        }
+
+        if y_scores is not None and len(np.unique(y_true)) > 1:
+            metrics["auc_roc"] = roc_auc_score(y_true, y_scores)
+
+        return metrics
+
+    @staticmethod
+    def get_confusion_matrix(
+        y_true: list | np.ndarray,
+        y_pred: list | np.ndarray,
+    ) -> np.ndarray:
+        """Return confusion matrix as a 2×2 numpy array."""
+        return confusion_matrix(y_true, y_pred)
+
+    @staticmethod
+    def get_classification_report(
+        y_true: list | np.ndarray,
+        y_pred: list | np.ndarray,
+        target_names: list = None,
+    ) -> str:
+        """Return a formatted classification report."""
+        if target_names is None:
+            target_names = ["Normal", "Abnormal"]
+        return classification_report(
+            y_true, y_pred, target_names=target_names, zero_division=0
+        )
+
+    @staticmethod
+    def format_metrics(metrics: dict) -> str:
+        """Pretty-print metrics to a formatted string."""
+        lines = []
+        for name, value in metrics.items():
+            if isinstance(value, float):
+                lines.append(f"  {name:>15s}: {value:.4f}")
+            else:
+                lines.append(f"  {name:>15s}: {value}")
+        return "\n".join(lines)