How to use latent variables to achieve downstream tasks

[735422713]

Hello. I'm having some trouble.
I've implemented training and testing with custom data, but I want to extract latent variables so that I can implement relevant downstream tasks such as classification, regression, etc.
Can you give me some advice on how to extract latent variables?
Thanks!!!

[clementchadebec]

Hello, sorry for the late reply.

You can use the embed method to get the corresponding latent variables.

benchmark_VAE/src/pythae/models/base/base_model.py

Lines 119 to 128 in bb785de

	def embed(self, inputs: torch.Tensor) -> torch.Tensor:
	"""Return the embeddings of the input data.
	Args:
	inputs (torch.Tensor): The input data to be embedded, of shape [B x input_dim].
	Returns:
	torch.Tensor: A tensor of shape [B x latent_dim] containing the embeddings.
	"""
	return self(DatasetOutput(data=inputs)).z

I hope this helps.

Best,

Clément