e8c1297dd6196fb486bfe7c6721041ff50ee1a47,matchzoo/models/drmm.py,DRMM,build,#DRMM#,40

Before Change

        // shape = [B, L, D]
        embed_query = embedding(query)
        // shape = [B, L]
        atten_mask = K.any(K.not_equal(query, self._params["mask_value"]),
                           axis=-1, keepdims=True)
        atten_mask = K.cast(atten_mask, K.floatx())
        atten_mask = K.expand_dims(atten_mask, axis=2)
        // shape = [B, L, D]
        attention_probs = self.attention_layer(embed_query, atten_mask)

        // Process right input.
        // shape = [B, L, 1]
        dense_output = self._make_multi_layer_perceptron_layer()(match_hist)

        // shape = [B, 1, 1]
        dot_score = keras.layers.Dot(axes=[1, 1])(
            [attention_probs, dense_output])

        flatten_score = keras.layers.Flatten()(dot_score)

        x_out = self._make_output_layer()(flatten_score)
        self._backend = keras.Model(inputs=[query, match_hist], outputs=x_out)

    @classmethod

After Change

        // shape = [B, L, D]
        embed_query = embedding(query)
        // shape = [B, L]
        atten_mask = K.not_equal(query, self._params["mask_value"])
        // shape = [B, L]
        atten_mask = K.cast(atten_mask, K.floatx())
        // shape = [B, L, D]
        atten_mask = K.expand_dims(atten_mask, axis=2)
        // shape = [B, L, D]
        attention_probs = self.attention_layer(embed_query, atten_mask)

        // Process right input.
        // shape = [B, L, 1]
        dense_output = self._make_multi_layer_perceptron_layer()(match_hist)

        // shape = [B, 1, 1]
        dot_score = keras.layers.Dot(axes=[1, 1])(
            [attention_probs, dense_output])

        flatten_score = keras.layers.Flatten()(dot_score)

        x_out = self._make_output_layer()(flatten_score)
        self._backend = keras.Model(inputs=[query, match_hist], outputs=x_out)

    @classmethod

Instances