ludwig/encoders/h3_encoders.py [164:204]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
            embeddings_on_cpu=embeddings_on_cpu,
            dropout=dropout,
            embedding_initializer=weights_initializer,
            embedding_regularizer=weights_regularizer
        )

        logger.debug('  FCStack')
        self.fc_stack = FCStack(
            layers=fc_layers,
            num_layers=num_fc_layers,
            default_fc_size=fc_size,
            default_use_bias=use_bias,
            default_weights_initializer=weights_initializer,
            default_bias_initializer=bias_initializer,
            default_weights_regularizer=weights_regularizer,
            default_bias_regularizer=bias_regularizer,
            default_activity_regularizer=activity_regularizer,
            # default_weights_constraint=weights_constraint,
            # default_bias_constraint=bias_constraint,
            default_norm=norm,
            default_norm_params=norm_params,
            default_activation=activation,
            default_dropout=dropout,
        )

    def call(
            self,
            inputs,
            training=None,
            mask=None
    ):
        """
            :param input_vector: The input vector fed into the encoder.
                   Shape: [batch x 19], type tf.int8
            :type input_vector: Tensor
            :param training: bool specifying if in training mode (important for dropout)
            :type training: bool
            :param mask: bool tensor encoding masked timesteps in the input
            :type mask: bool
         """
        input_vector = tf.cast(inputs, tf.int32)
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -



ludwig/encoders/date_encoders.py [206:247]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
            embeddings_on_cpu=embeddings_on_cpu,
            dropout=dropout,
            embedding_initializer=weights_initializer,
            embedding_regularizer=weights_regularizer
        )

        logger.debug('  FCStack')
        self.fc_stack = FCStack(
            layers=fc_layers,
            num_layers=num_fc_layers,
            default_fc_size=fc_size,
            default_use_bias=use_bias,
            default_weights_initializer=weights_initializer,
            default_bias_initializer=bias_initializer,
            default_weights_regularizer=weights_regularizer,
            default_bias_regularizer=bias_regularizer,
            default_activity_regularizer=activity_regularizer,
            # default_weights_constraint=weights_constraint,
            # default_bias_constraint=bias_constraint,
            default_norm=norm,
            default_norm_params=norm_params,
            default_activation=activation,
            default_dropout=dropout,
        )

    def call(
            self,
            inputs,
            training=None,
            mask=None
    ):
        """
            :param input_vector: The input vector fed into the encoder.
                   Shape: [batch x 19], type tf.int8
            :type input_vector: Tensor
            :param training: bool specifying if in training mode (important for dropout)
            :type training: bool
            :param mask: bool specifying masked values
            :type mask: bool
         """
        # ================ Embeddings ================
        input_vector = tf.cast(inputs, tf.int32)
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -