CODEWITHSUNDEEP
pythontensorflow
Yao Yingjie
Yao Yingjie

Reputation: 468

elegant way to set tensor value according to a constant python list

I have a const python list

listA = [1, 2, 3, 23, ...]

and a tensor

tensorA = [[1, 3, 5, 7, 23,...]]

Now I want to modify to tensorA according to listA:

for each element x in tensorA, if x also in listA then keep it as it is, otherwise use the default value just like -1.

After this transformation, tensorA goes like

tensorB = [[1, 3, -1, -1, 23, ...]]

Is there any elegant way to do this transformation?

Upvotes: 1

Views: 181

Answers (2)

javidcf
javidcf

Reputation: 59731

Since TensorFlow does not currently have anything like NumPy's isin, you would need to do an all-to-all comparison:

import tensorflow as tf

listA = tf.constant([1, 2, 3, 23])
tensorA = tf.constant([[1, 3, 5, 7, 23]])

isInList = tf.reduce_any(tf.equal(tf.expand_dims(tensorA, axis=-1), listA), axis=-1)
tensorB = tf.where(isInList, tensorA, -1)
tf.print(tensorB)
# [[1 3 -1 -1 23]]

Upvotes: 2

Jab
Jab

Reputation: 27515

A list comprehension?

tensorB = [[n if n in listA else -1 for n in tensorA]]

Upvotes: 0

Related Questions