e9aea97df1dc7878827ac193ba75cbea0b3ee351,ludwig/models/modules/sequence_decoders.py,SequenceGeneratorDecoder,__init__,#SequenceGeneratorDecoder#Any#Any#Any#Any#Any#Any#Any#Any#Any#Any#Any#Any#Any#Any#Any#Any#Any#Any#,32

Before Change

        else:
            self.vocab_size = self.num_classes

        self.embeddings_dec = Embedding(num_classes, embedding_size)
        self.decoder_cell = LSTMCell(state_size)

        if attention_mechanism:
            if attention_mechanism == "bahdanau":
                pass
            elif attention_mechanism == "luong":
                self.attention_mechanism = tfa.seq2seq.LuongAttention(
                    state_size,
                    None,  // todo tf2: confirm on need
                    memory_sequence_length=max_sequence_length  // todo tf2: confirm inputs or output seq length
                )
            else:
                raise ValueError(
                    "Attention specificaiton "{}" is invalid.  Valid values are "
                    ""bahdanau" or "luong".".format(self.attention_mechanism))

            self.decoder_cell = tfa.seq2seq.AttentionWrapper(
                self.decoder_cell,
                self.attention_mechanism
            )

        self.sampler = tfa.seq2seq.sampler.TrainingSampler()

        self.projection_layer = Dense(
            units=num_classes,
            use_bias=use_bias,
            kernel_initializer=weights_initializer,
            bias_initializer=bias_initializer,
            kernel_regularizer=weights_regularizer,
            bias_regularizer=bias_regularizer,
            activity_regularizer=activity_regularizer
        )

        self.decoder = \
            tfa.seq2seq.basic_decoder.BasicDecoder(self.decoder_cell,
                                                    self.sampler,

After Change

        self.decoder_embedding = tf.keras.layers.Embedding(
            input_dim=output_vocab_size,
            output_dim=embedding_dims)
        self.dense_layer = tf.keras.layers.Dense(output_vocab_size)
        self.decoder_rnncell = tf.keras.layers.LSTMCell(rnn_units)

        // Sampler
        self.sampler = tfa.seq2seq.sampler.TrainingSampler()

        self.attention_mechanism = None
        self.rnn_units = rnn_units

        print("setting up attention for", attention_mechanism)
        if attention_mechanism is not None:
            self.attention_mechanism = self.build_attention_mechanism(
                attention_mechanism,
                dense_units
            )
            self.decoder_rnncell = self.build_rnn_cell()

        self.decoder = tfa.seq2seq.BasicDecoder(self.decoder_rnncell,
                                                sampler=self.sampler,
                                                output_layer=self.dense_layer)

Instances