Resolved log(0) error in KL divergence TheAlgorithms#12233

Author: anscian

machine_learning/loss_functions.py

@@ -659,7 +659,9 @@ def kullback_leibler_divergence(y_true: np.ndarray, y_pred: np.ndarray) -> float
     if len(y_true) != len(y_pred):
         raise ValueError("Input arrays must have the same length.")
 
-    kl_loss = y_true * np.log(y_true / y_pred)
+    kl_loss = np.concatenate((y_true[None, :], y_pred[None, :]))
+    kl_loss = kl_loss[:, ~np.any(kl_loss == 0, axis=0)]
+    kl_loss = kl_loss[0] * np.log(kl_loss[0] / kl_loss[1])
     return np.sum(kl_loss)