Broadcasting during `scatter_nd()`?

TensorFlow General Discussion
1

Let’s say I have an array (final) of shape (2048, 128). I have got another array called indices of shape (2048, 1).

If I wanted to scatter a value of 1 in final w.r.t indices, how will I do that?

What I have tried:

indices = tf.constant(np.random.randint(1000, size=(2048, 1)))
output = tf.scatter_nd(indices, 1, [2048, 128])

Results into:

InvalidArgumentError: Updates shape must have rank at least one. Found:[] [Op:ScatterNd]

What is the recommended solution?

3

Oh, I guess I can just do:

indices = tf.constant(np.random.randint(1000, size=(2048, 1)))
output = tf.one_hot(indices, 128)