CODEWITHSUNDEEP
pythontensorflow
Toby Mao
Toby Mao

Reputation: 115

Tensorflow, how to generate all-zero tensor shaped by a placeholder

I wanted to use the tf.where function in tensorflow.

selected_data = tf.where(mask,some_place_holder,zeros)

however, when I wrote

zeros = tf.zeros(some_place_holder.shape)

error occurs:

ValueError: Cannot convert a partially known TensorShape to a Tensor: (?, 1000, 10)

I also tried to use tf.fill, but similar errors occured.

Well, there indeed some solution such as

zeros = tf.matmul(some_place_holder , tf.zeros([some_place_holder.shape[-1],some_place_holder.shape[-1]]))

but is there any better solution?

Upvotes: 0

Views: 1057

Answers (1)

akuiper
akuiper

Reputation: 214927

You can use tf.zeros_like(some_place_holder):

input_tensor = tf.placeholder(tf.int8, shape=[None, 3])

zeros = tf.zeros_like(input_tensor)

with tf.Session() as sess:
    print(sess.run(zeros, feed_dict={input_tensor: [[1,2,3]]}))
# [[0 0 0]]

Upvotes: 2

Related Questions