Theano tensor slicing… how to use boolean to slice?

Question

In numpy, if I have a boolean array, I can use it to select elements of another array:

>>> import numpy as np
>>> x = np.array([1, 2, 3])
>>> idx = np.array([True, False, True])
>>> x[idx] 
array([1, 3])

I need to do this in theano. This is what I tried, but I got an unexpected result.

>>> from theano import tensor as T
>>> x = T.vector()
>>> idx = T.ivector()
>>> y = x[idx]
>>> y.eval({x: np.array([1,2,3]), idx: np.array([True, False, True])})
array([ 2.,  1.,  2.])

Can someone explain the theano result and suggest how to get the numpy result? I need to know how to do this in order to properly instantiate a 'givens' argument in a theano function declaration. Thanks in advance.

Answer 1

This is not supported in theano:

We do not support boolean masks, as Theano does not have a boolean type (we use int8 for the output of logic operators).

Theano indexing with a “mask” (incorrect approach):

>>> t = theano.tensor.arange(9).reshape((3,3))
>>> t[t > 4].eval()  # an array with shape (3, 3, 3)
...

Getting a Theano result like NumPy:

>>> t[(t > 4).nonzero()].eval()
array([5, 6, 7, 8])

So you need y = x[idx.nonzero()]