b9b992e52fb42f137e6aa0de6e9f201ef0140571,basic/model.py,,bi_attention,#Any#Any#Any#Any#Any#Any#Any#Any#,16

Before Change

        if tensor_dict is not None:
            a_h = tf.nn.softmax(h_logits)  // [N, M, JX]
            a_u = tf.nn.softmax(u_logits)  // [N, M, JX, JQ]
            tensor_dict["a_h"] = a_h
            tensor_dict["a_u"] = a_u
        return u_avg, h_a, u_a

After Change

            and_mask = h_mask_aug & u_mask_aug
            u_avg = tf.reduce_sum(u_aug * tf.cast(tf.expand_dims(u_mask_aug, -1), "float"), 3)

        if config.sh:
            cell = SHCell(h.get_shape()[3])
            h_len = tf.reduce_sum(tf.cast(h_mask, "int32"), 2)  // [N, M, JX]
            in_ = tf.concat(3, [h, u_avg])
            (fw_h, bw_h), _ = bidirectional_dynamic_rnn(cell, cell, in_, h_len, dtype="float", scope="u1")  // [N, M, JX, 2d]
            h_a = fw_h + bw_h

        else:
            h_logits = get_logits([h, u_avg], None, True, wd=config.wd, mask=h_mask,
                                  is_train=is_train, func="mul_linear", scope="h_logits")  // [N, M, JX]
            h_a = softsel(h, h_logits)  // [N, M, d]
            h_a_tiled = tf.tile(tf.expand_dims(h_a, 2), [1, 1, JX, 1])

        u_logits = get_logits([h_aug, u_aug], None, True, wd=config.wd, mask=and_mask,
                              is_train=is_train, func="mul_linear", scope="u_logits")  // [N, M, JX, JQ]
        u_a = softsel(u_aug, u_logits)  // [N, M, JX, d]
        if tensor_dict is not None:

Instances