maybe_mask_tensor

bayesflow.utils.maybe_mask_tensor(data: Tensor, mask: Tensor | None = None, replacement: Tensor = None) → Tensor

Apply a binary mask to a tensor if a mask is passed, blending with a replacement where masked.

Parameters:

dataTensor

The tensor to mask.

maskTensor or None, optional

Binary mask where 1.0 = keep, 0.0 = replace. If None, data is returned unchanged.

replacementTensor, optional

Values to use where the mask is 0. If None, zeros are used.

Returns:

Tensor

mask * data + (1 - mask) * replacement, or data when mask is None.