BaseEncoder

class pyraug.models.nn.BaseEncoder

This is a base class for Encoders neural networks.

forward(x)

This function must be implemented in a child class. It takes the input data and returns (mu, log_var) in that order. If you decide to provide your own encoder network, you must make your model inherit from this class by setting and the define your forward function as such:

class My_Encoder(BaseEncoder):

    def __init__(self):
        BaseEncoder.__init__(self)
        # your code

    def forward(self, x):
        # your code
        return mu, log_var

Parameters

x (torch.Tensor) – The input data that must be encoded

Returns

The mean \(\mu_{\phi}\) and the log of the variance ():math:log Sigma) of the approximate distribution. As is common :Sigma is diagonal and so only the diagonal coefficients should be output

Return type

(Tuple[torch.Tensor, torch.Tensor)

Warning

The output order in here important. Do not forget to set \(\mu\) as first argument and the log variance then.