Dimension must equal, MatMul with input shapes: [?,53],[150,200]

Question

我想将两个不同维度的向量送进Bi-LSTM中，然后concat起来，但是报错。

with tf.variable_scope("bi-lstm"):
    cell_bw = LSTMCell(self.hidden_dim)
    cell_fw = LSTMCell(self.hidden_dim)

    (output_fw_seq_e, output_bw_seq_e), _ = tf.nn.bidirectional_dynamic_rnn(
        cell_fw=cell_fw,
        cell_bw=cell_bw,
        inputs=self.word_embeddings,
        sequence_length=self.sequence_lengths,
        dtype=tf.float32)
    output_e = tf.concat([output_fw_seq_e, output_bw_seq_e], axis=-1)

    (output_fw_seq_d, output_bw_seq_d), _ = tf.nn.bidirectional_dynamic_rnn(
        cell_fw=cell_fw,
        cell_bw=cell_bw,
        inputs=self.features_dim,
        sequence_length=self.sequence_lengths,
        dtype=tf.float32)
    output_d = tf.concat([output_fw_seq_d, output_bw_seq_d], axis=-1)

    output = tf.concat([output_e, output_d], axis=-1)
    output = tf.nn.dropout(output, self.dropout_pl)

self.hidden_dim: 50维
self.word_embeddings: 100维
self.features: 3维

求助！
补充Weight和bias代码如下:

with tf.variable_scope("proj"):
    W = tf.get_variable(name="W",
                        shape=[2 * self.hidden_dim, self.num_tags],
                        initializer=tf.contrib.layers.xavier_initializer(),
                        dtype=tf.float32)

    b = tf.get_variable(name="b",
                        shape=[self.num_tags],
                        initializer=tf.zeros_initializer(),
                        dtype=tf.float32)

    s = tf.shape(output)
    output = tf.reshape(output, [-1, 2 * self.hidden_dim])
    pred = tf.matmul(output, W) + b

Answer 1

[?,53]跟[150,200]不能矩阵乘，第一个矩阵的第二维和第二个矩阵的第一维要相同。
问题应该是上一步的输出不对，或者模型不对。

试一下生成网络结构图，效果如下，http://ethereon.github.io/net...