TensorFlow 2.0: How to update tensors?

Question

In TensorFlow 1.x, to update a tensor, I would use tf.scatter_update , to only update the relevant part of the tensor.

How can we do the same thing in TF 2.0?

Answer 1

You can use tf.tensor_scatter_nd_update() :

import tensorflow as tf
import numpy as np 

tensor = tf.convert_to_tensor(np.ones((2, 2)), dtype=tf.float32)
indices = tf.constant([[0, 0]])
updates = tf.constant([0.0])

tf.tensor_scatter_nd_update(tensor, indices, updates).numpy()
# array([[0., 1.],
#        [1., 1.]], dtype=float32)