python - 如何连接 "Jagged"张量

Question

我正在尝试编写 this 的实现我在 TensorFlow 的论文中遇到了一些障碍。在我的池化层中，我必须将所有内容连接在一起。这是我使用的代码:

    pooled_outputs = []
    for i, filter_size in enumerate(filter_sizes):
        with tf.name_scope("conv-maxpool-%s" % filter_size):
            # Conv layer
            filter_shape = [filter_size, embedding_size, 1, num_filters]
            # W is the filter matrix
            W = tf.Variable(tf.truncated_normal(filter_shape, stddev=0.1), name="W")
            b = tf.Variable(tf.constant(0.1, shape=[num_filters]), name="b")
            conv = tf.nn.conv2d(
                self.embedded_chars_expanded,
                W,
                strides=[1, 1, 1, 1],
                padding="VALID",
                name="conv"
            )

            # Apply nonlinearity
            h = tf.nn.relu(tf.nn.bias_add(conv, b), name="relu")

            # Max-pooling layer over the outputs
            pooled = tf.nn.max_pool(
                h,
                ksize=[1, sequence_lengths[i] - filter_size + 1, 1, 1],
                strides=[1, 1, 1, 1],
                padding="VALID",
                name="pool"
            )
            pooled_outputs.append(pooled)

    # Combine all of the pooled features
    num_filters_total = num_filters * len(filter_sizes)

    print(pooled_outputs)
    pooled_outputs = [tf.reshape(out, ["?", 94, 1, self.max_length]) for out in pooled_outputs] # The problem line

    self.h_pool = tf.concat(3, pooled_outputs)

当我运行此代码时，它会为 pooled_outputs 打印出以下内容:

[<tf.Tensor 'conv-maxpool-3/pool:0' shape=(?, 94, 1, 128) dtype=float32>, <tf.Tensor 'conv-maxpool-4/pool:0' shape=(?, 51, 1, 128) dtype=float32>, <tf.Tensor 'conv-maxpool-5/pool:0' shape=(?, 237, 1, 128) dtype=float32>]

我最初尝试这段代码时没有使用 pooled_outputs = [tf.reshape(out, ["?", 94, 1, self.max_length]) for out in pooled_outputs] 行，我收到此错误:

ValueError: Dimension 1 in both shapes must be equal, but are 51 and 237

当我添加 reshape 线时，出现此错误:

TypeError: Expected binary or unicode string, got 94

我知道的第二个错误是因为我传递了一个“？”对于新的大小，我认为第一个错误是因为张量大小不同。 我怎样才能正确地填充这些张量，以便我可以毫无问题地连接它们？

Answer 1

您可以将 -1 作为形状的组成部分之一传递给 tf.reshape 方法；它将根据张量的形状自动推断，因此总大小将相同。

所以，尝试将问题行更改为

pooled_outputs = [tf.reshape(out, [-1, 94, 1, self.max_length]) for out in pooled_outputs]

请参阅documentation了解详情