the EncoderProcessorDecoder forward types

[marjanfamili]

The encoder outputs the type of TensorBMStarL :

    def encode(self, x: Batch) -> TensorBMStarL:
        return self.encoder_decoder.encoder(x)

The forwards pass of the EncoderProcessorDecoder is :

    def forward(self, batch: Batch) -> TensorBTSPlusC:
        return self.decode(self.processor(self.encode(batch)))

however the processor should expect EncodedBatch, Is this all correct ?

[sgreenbury]

Just adding here too to catpture discussion: thanks for mentioning this - the EncoderProcessorDecoder's forward pass aims to do a full pass from batch to encoded tensor to mapped encoded tensor to decoded tensor:

     def forward(self, batch: Batch) -> TensorBTSPlusC:
        # return self.decode(self.processor(self.encode(batch)))
        encoded = self.encoder_decoder.encode(batch)
        processed = self.processor.map(encoded)
        decoded = self.encoder_decoder.decoder.decode(processed)
        return decoded

but the current __call__s in the base classes are not necessarily calling these methods correctly in the current implementation so this is to be checked.