I'm trying to do something like this:
Suppose the input Tensor is a (2, 3) Tensor with the value like:
[[1 2]
[3 4]
[5 6]]
I am flattening the 2D Tensor into 1D by doing:
input = tf.reshape(input, [-1])
So the input now becomes [1 2 3 4 5 6].
But I also create a 1D Tensor that indexes the previous Tensor, so the desired output is [0 0 1 1 2 2]. How should I create this Tensor in TF?
In general, if the input Tensor has a shape of (X, Y). I would like to create a 1D Tensor that looks like:
[0 0 0 ... 0 1 ....1 2 ... 2 ... X-1 ... X-1]
where each value is repeated Y - 1 times.