MAE percentage metrics used

utils/metrics.py

@@ -1,37 +1,81 @@
-import numpy as np 
-
-
-# Adding type hints for better code clarity and numpy style comments for documentation  
-def mae(y_true: np.ndarray, y_pred: np.ndarray) -> float:
-    """Mean Absolute Error
-    Parameters
-    ----------
-    y_true : np.ndarray
-        True values
-    y_pred : np.ndarray
-        Predicted values
-    Returns
-    -------
-    float
-        Mean Absolute Error
+# import numpy as np 
+
+
+# # Adding type hints for better code clarity and numpy style comments for documentation  
+# def mae(y_true: np.ndarray, y_pred: np.ndarray) -> float:
+#     """Mean Absolute Error
+#     Parameters
+#     ----------
+#     y_true : np.ndarray
+#         True values
+#     y_pred : np.ndarray
+#         Predicted values
+#     Returns
+#     -------
+#     float
+#         Mean Absolute Error
+#     """
+#     y_true = np.array(y_true)
+#     y_pred = np.array(y_pred)
+#     return np.mean(np.abs(y_true - y_pred))
+
+# def bias(y_true: np.ndarray, y_pred: np.ndarray) -> float:
+#     """Bias
+#     Parameters
+#     ----------
+#     y_true : np.ndarray
+#         True values
+#     y_pred : np.ndarray
+#         Predicted values
+#     Returns
+#     -------
+#     float
+#         Bias
+#     """
+#     y_true = np.array(y_true)
+#     y_pred = np.array(y_pred)
+#     return np.mean(y_pred - y_true)
+
+
+import numpy as np
+
+def mae(y_true: np.ndarray, y_pred: np.ndarray, eps: float = 1e-12) -> float:
+    """
+    Sum-based MAE percentage, in %.
+
+    Formula:
+        MAE%_sum = 100 * sum(|y_true - y_pred|) / sum(|y_true|)
+
+    Notes:
+    - Equivalent to WAPE when computed on one series/window.
+    - Stable for intermittent demand compared with per-point MAPE.
     """
-    y_true = np.array(y_true)
-    y_pred = np.array(y_pred)
-    return np.mean(np.abs(y_true - y_pred))
-
-def bias(y_true: np.ndarray, y_pred: np.ndarray) -> float:
-    """Bias
-    Parameters
-    ----------
-    y_true : np.ndarray
-        True values
-    y_pred : np.ndarray
-        Predicted values
-    Returns
-    -------
-    float
-        Bias
+    y_true = np.asarray(y_true, dtype=float)
+    y_pred = np.asarray(y_pred, dtype=float)
+
+    denom = np.sum(np.abs(y_true))
+    if denom < eps:
+        return np.nan  # undefined if all actuals are ~0
+
+    return 100.0 * np.sum(np.abs(y_true - y_pred)) / denom
+
+
+def bias(y_true: np.ndarray, y_pred: np.ndarray, eps: float = 1e-12) -> float:
     """
-    y_true = np.array(y_true)
-    y_pred = np.array(y_pred)
-    return np.mean(y_pred - y_true)
+    Sum-based Bias percentage, in %.
+
+    Formula:
+        Bias%_sum = 100 * sum(y_pred - y_true) / sum(|y_true|)
+
+    Sign:
+        > 0  => over-forecasting in aggregate
+        < 0  => under-forecasting in aggregate
+    """
+    y_true = np.asarray(y_true, dtype=float)
+    y_pred = np.asarray(y_pred, dtype=float)
+
+    denom = np.sum(np.abs(y_true))
+    if denom < eps:
+        return np.nan  # undefined if all actuals are ~0
+
+    return 100.0 * np.sum(y_pred - y_true) / denom