Two architectures: the GAN and the recurrent VAE (LSTM) don't seem to work. GAN never converges and RNN just generates garbage. Also, I have not been successful in using the BLOSUM encoding of the data. Not sure what the correct solution to this is, but my solution has so far been to just always use the fully connected VAE and the categorical encoding.
Two architectures: the GAN and the recurrent VAE (LSTM) don't seem to work. GAN never converges and RNN just generates garbage. Also, I have not been successful in using the BLOSUM encoding of the data. Not sure what the correct solution to this is, but my solution has so far been to just always use the fully connected VAE and the categorical encoding.