CondVAEModel.decode

CondVAEModel.decode(y: torch.Tensor, z: array | torch.Tensor | None = None, z_std: float = 1.0) → torch.Tensor

Decode the latent representation into the original data space. Extends the parent’s decode function by sampling z from a normal distribution if it is not provided.

Parameters:

• y (torch.Tensor) – The conditional data.

• z (Union[np.array, torch.Tensor], optional, default=None) – The latent representation to decode.

• z_std (float, optional, default=1.0) – Used only if z is None. Defines the standard deviation of the Gaussian with which z will be sampled. Controls diversity vs. reliability.

Returns:

torch.Tensor – A tensor containing the reconstructed (generated) data.