Tensorflow重塑张量

Question

我有一个预测张量(实际网络)

(Pdb) pred
<tf.Tensor 'transpose_1:0' shape=(?, 200, 200) dtype=float32>

和张力

y = tf.placeholder("float", [None, n_steps, n_classes])

(Pdb) y
<tf.Tensor 'Placeholder_1:0' shape=(?, 200, 200) dtype=float32>

我想把它喂进去

f.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(pred, y))

但是,它需要尺寸 [batch_size, num_classes]

所以我想重塑两者pred,y使它们看起来像这样

<tf.Tensor 'transpose_1:0' shape=(?, 40000) dtype=float32>

但是当我跑步时reshape我得到了

(Pdb) tf.reshape(pred, [40000])
<tf.Tensor 'Reshape_1:0' shape=(40000,) dtype=float32>

而不是(?,40000)如何保持这个None维度？(批量大小)

我还发布了所有相关代码......

# tf Graph input
x = tf.placeholder("float", [None, n_steps, n_input])
y = tf.placeholder("float", [None, n_steps, n_classes])


# Define weights
weights = {
    'hidden': tf.Variable(tf.random_normal([n_hidden, n_classes]), dtype="float32"),
    'out': tf.Variable(tf.random_normal([n_hidden, n_classes]), dtype="float32")
}
biases = {
    'hidden': tf.Variable(tf.random_normal([n_hidden]), dtype="float32"),
    'out': tf.Variable(tf.random_normal([n_classes]), dtype="float32")
}


def RNN(x, weights, biases):

    # Prepare data shape to match `rnn` function requirements
    # Current data input shape: (batch_size, n_steps, n_input)
    # Permuting batch_size and n_steps
    x = tf.transpose(x, [1, 0, 2])
    # Reshaping to (n_steps*batch_size, n_input)

    x = tf.reshape(x, [-1, n_input])
    # Split to get a list of 'n_steps' tensors of shape (batch_size, n_hidden)
    # This input shape is required by `rnn` function
    x = tf.split(0, n_steps, x)
    # Define a lstm cell with tensorflow
    lstm_cell = rnn_cell.BasicLSTMCell(n_hidden, forget_bias=1.0, state_is_tuple=True)
    outputs, states = rnn.rnn(lstm_cell, x, dtype=tf.float32)
    output_matrix = []

    for i in xrange(n_steps):
        temp = tf.matmul(outputs[i], weights['out']) + biases['out']
        # temp = tf.matmul(weights['hidden'], outputs[i]) + biases['hidden']
        output_matrix.append(temp)
    pdb.set_trace()

    return output_matrix

pred = RNN(x, weights, biases)
# temp = RNN(x)
# pdb.set_trace()
# pred = tf.shape(temp)
pred = tf.pack(tf.transpose(pred, [1,0,2]))
cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(pred, y))

Answer 1

我是雅罗斯拉夫评论中另一个问题的答案之一的作者。您可以将-1用于None维度。

您可以使用tf.reshape（）轻松完成此操作，而无需知道批处理大小。

x = tf.placeholder(tf.float32, shape=[None, 9,2])
shape = x.get_shape().as_list()        # a list: [None, 9, 2]
dim = numpy.prod(shape[1:])            # dim = prod(9,2) = 18
x2 = tf.reshape(x, [-1, dim])           # -1 means "all"

无论运行时批处理大小如何，最后一行中的-1表示整个列。您可以在tf.reshape（）中看到它。