Question

我正在尝试将张量从2d调整为1d，但是当我尝试时，我遇到了一个非常奇怪的错误，即2、2无法调整为4。我做错了什么？ >

import tensorflow as tf

x_size = 2
y_size = 2
batch_size = 10

sess = tf.InteractiveSession()

x = tf.placeholder(tf.float32, (batch_size, x_size, y_size))
y_ = tf.placeholder(tf.float32, (None, 1))

x = tf.reshape(x, (batch_size, x_size * y_size))
y = tf.layers.dense(x, 1)

loss = tf.reduce_mean(tf.losses.mean_squared_error(y, y_))

a = [[list(range(y_size)) for i in range(x_size)] for i in range(batch_size)]
b = [[1] for i in range(batch_size)]
sess.run(loss, feed_dict={x: a, y_: b})

这给了我

ValueError: Cannot feed value of shape (10, 2, 2) for Tensor 'Reshape:0', which has shape '(10, 4)'

Answer 1

错误是因为我将x传递到feed_dict中，但是我用重塑形状覆盖了x。通过将第12和13行替换为

nn = tf.reshape(x, (batch_size, x_size * y_size))
y = tf.layers.dense(nn, 1)

并添加

sess.run(tf.global_variables_initializer())

我能够使其完美运行。讲故事的寓意，请始终将占位符分开。

Answer 2

您定义的x张量实际上是等级2（3d）设置为[10，2，2]。那么，假设您想使用2D，那么

这只是黑暗中的镜头，个人还没有进行重塑，但是在示例中，“ flatten”参数为-1？

因此，从API doc复制就可以了：

# tensor 't' is [[[1, 1, 1],
#                 [2, 2, 2]],
#                [[3, 3, 3],
#                 [4, 4, 4]],
#                [[5, 5, 5],
#                 [6, 6, 6]]]    
# pass '[-1]' to flatten 't'
reshape(t, [-1]) ==> [1, 1, 1, 2, 2, 2, 3, 3, 3, 4, 4, 4, 5, 5, 5, 6, 6, 6]

# -1 can also be used to infer the shape

# -1 is inferred to be 9:
reshape(t, [2, -1]) ==> [[1, 1, 1, 2, 2, 2, 3, 3, 3],
                     [4, 4, 4, 5, 5, 5, 6, 6, 6]]
# -1 is inferred to be 2:
reshape(t, [-1, 9]) ==> [[1, 1, 1, 2, 2, 2, 3, 3, 3],
                     [4, 4, 4, 5, 5, 5, 6, 6, 6]]

因此，如果想从[10，2，2]变为[10，4]，从理论上讲这行不通吗？

x = tf.reshape(x, [batch_size, -1])

或者，如果您实际上想一直下降到一维，则：

x = tf.reshape(x, [-1])

形状正确的Tensorflow调整大小失败

2 个答案: