Beamsearch.py

  # Performs a beam search, to be run in the "Decoder"
  def Beamsearch(self,
             predict_fun, # RNN() with: new_states as a list, outputs of shape (batch_size, n_symbols)
             states0,     # RNN()'s initial layer states are equal to states0
             batch_size,  # Size derived from initial layer states0
             start_symbol # Tagging the encoded target sequence with <S> as a start symbol for example
             stop_symbol  # '                                      ' </S> as a stop symbol for example
             max_length,  # Maximum length of a given sequence
             beam_size=4  # See [simple beam search](https://github.com/SeitaroShinagawa/simple_beamsearch) for an example of probabilities
            ):
      return (sequence, mask, scores)