Sequence-based indexing in theano is deprecated

[timokau]

In csrank/discretechoice/nested_logit_model.py we make use of sequence indexing in theano multiple times. For example:

cs-ranking/csrank/discretechoice/nested_logit_model.py

Line 175 in ba03234

rows, cols = tt.eq(self.y_nests, i).nonzero()

Here rows and cols are both 1d tensors which are then used to index a different tensor:

cs-ranking/csrank/discretechoice/nested_logit_model.py

Line 176 in ba03234

utility = tt.set_subtensor(utility[rows, cols], tt.dot(self.Xt[rows, cols], weights[i]))

Theano complains that this is deprecated:

FutureWarning: Using a non-tuple sequence for multidimensional indexing is deprecated; use `arr[tuple(seq)]` instead of `arr[seq]`. In the future this will be interpreted as an array index, `arr[np.array(seq)]`, which will result either in an error or a different result.

I'm not sure how to fix this. According to the documentation, theano should support boolean mask indexing. So I thought we should be able to do

mask = tt.eq(self.y_nests, i)
utility = tt.set_subtensor(utility[mask], tt.dot(self.Xt[mask], weights[i]))

instead (as tt.eq should return a boolean mask). But unfortunately that doesn't work; it gives the same warning.

Any ideas?

[kiudee]

I think this problem was caused by a new version of NumPy and has been discussed here:
Theano/Theano#6703

[timokau]

Right, so the api is still supported by theano. I've confirmed that the issue is fixed in upstream by running the testsuite with a patched theano. Another case of "not our fault" then.

[timokau]

For us, the commits 082yqypw8vddxly47p26l9sgda3z598i098dldp2d0k04vawlblx and 0y2g08arbcffgkd9w0aan1g2yn7n06ma8qwd8i2vvbj5z1n28xvq are interesting (not included in the PR you linked, but similar in intention).