Merge pull request #74 from aertslab/dev

update msecosine log loss function

src/crested/tl/losses/_cosinemse_log.py

@@ -30,6 +30,8 @@ class CosineMSELogLoss(keras.losses.Loss):
         Name of the loss function.
     reduction
         Type of reduction to apply to loss.
+    multiplier
+        Scalar to multiply the predicted value with. When predicting mean coverage, multiply by 1000 to get actual count. Keep to 1 when predicting insertion counts.
 
     Notes
     -----
@@ -50,11 +52,13 @@ def __init__(
         max_weight: float = 1.0,
         name: str | None = "CosineMSELogLoss",
         reduction: str = "sum_over_batch_size",
+        multiplier: float = 1000,
     ):
         """Initialize the loss function."""
         super().__init__(name=name)
         self.max_weight = max_weight
         self.reduction = reduction
+        self.multiplier = multiplier
 
     def call(self, y_true, y_pred):
         """Compute the loss value."""
@@ -64,13 +68,13 @@ def call(self, y_true, y_pred):
         y_true1 = keras.utils.normalize(y_true, axis=-1)
         y_pred1 = keras.utils.normalize(y_pred, axis=-1)
 
-        log_y_pred_pos = keras.ops.log(1 + 1000 * keras.ops.maximum(y_pred, 0))
+        log_y_pred_pos = keras.ops.log(1 + self.multiplier * keras.ops.maximum(y_pred, 0))
         log_y_pred_neg = -keras.ops.log(
-            1 + keras.ops.abs(1000 * keras.ops.minimum(y_pred, 0))
+            1 + keras.ops.abs(self.multiplier * keras.ops.minimum(y_pred, 0))
         )
 
         log_y_pred = log_y_pred_pos + log_y_pred_neg
-        log_y_true = keras.ops.log(1 + 1000 * y_true)
+        log_y_true = keras.ops.log(1 + self.multiplier * y_true)
 
         mse_loss = keras.ops.mean(keras.ops.square(log_y_pred - log_y_true))
         weight = keras.ops.abs(mse_loss)