What is the CNTK way to convert a vector that contains labels as indices to a one hot-representation?

问题

In CNTK I need a way to convert a vector that contains labels as indices (just a regular vector, not a sparse representation) to a one hot-representation.

Here is an example for 5 classes:

Input

[2, 0, 1, 1]

Desired output:

[[0,0,1,0,0],
[1,0,0,0,0],
[0,1,0,0,0],
[0,1,0,0,0]]

Is there a way without going through Python/numpy?

回答1:

The ‘Value.one_hot’ method does this (converts to a sparseCSC matrix representation internally).

https://www.cntk.ai/pythondocs/cntk.html?highlight=one_hot#cntk.core.Value.one_hot

回答2:

If you're going to use these labels as tags for output layer then probably you'll have to use a numpy array. You should look at this answer.

来源：https://stackoverflow.com/questions/42896773/what-is-the-cntk-way-to-convert-a-vector-that-contains-labels-as-indices-to-a-on