Theano: FutureWarning: Using a non-tuple sequence for multidimensional indexing is deprecated

[lucianopaz]

While using pymc3 I ran into some FutureWarning's while performing advanced indexing on a tensor variable. This is a minimal code that reproduces the warning (treated as an exception) along with its traceback.

import warnings
import numpy as np
import theano
from theano import tensor


theano.config.compute_test_value = 'raise'

with warnings.catch_warnings(record=True) as w:
    # Cause all warnings to always be triggered.
    warnings.simplefilter("error")
    n = 10
    size = 1000

    a = np.random.randn(n)
    inds = np.random.randint(0, n, size=size)

    A = tensor.dvector('A')
    A.tag.test_value = a
    B = A[inds, None]

The traceback is the following

  File "/home/lpaz/Dropbox/Luciano/issue pymc3/theano_warn.py", line 28, in <module>
    B = A[inds, None]

  File "/home/lpaz/anaconda3/lib/python3.6/site-packages/theano/tensor/var.py", line 572, in __getitem__
    return theano.tensor.subtensor.advanced_subtensor(self, *args)

  File "/home/lpaz/anaconda3/lib/python3.6/site-packages/theano/gof/op.py", line 674, in __call__
    required = thunk()

  File "/home/lpaz/anaconda3/lib/python3.6/site-packages/theano/gof/op.py", line 892, in rval
    r = p(n, [x[0] for x in i], o)

  File "/home/lpaz/anaconda3/lib/python3.6/site-packages/theano/tensor/subtensor.py", line 2190, in perform
    rval = inputs[0].__getitem__(inputs[1:])

 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.

[nouiz]

Which version of NumPy do you use?

I suppose a new version of NumPy is breaking stuff.
But I do not understand why they tell this should be a warning, but it
break Theano.

On Tue, Dec 11, 2018 at 9:10 AM Luciano Paz [e-mail]
wrote:

While using pymc3 I ran into some FutureWarning's while performing
advanced indexing on a tensor variable. This is a minimal code that
reproduces the warning (treated as an exception) along with its traceback.

import warningsimport numpy as npimport theanofrom theano import tensor

theano.config.compute_test_value = 'raise'
with warnings.catch_warnings(record=True) as w:
# Cause all warnings to always be triggered.
warnings.simplefilter("error")
n = 10
size = 1000

a = np.random.randn(n)
inds = np.random.randint(0, n, size=size)

A = tensor.dvector('A')
A.tag.test_value = a
B = A[inds, None]

The traceback is the following

File "/home/lpaz/Dropbox/Luciano/issue pymc3/theano_warn.py", line 28, in
B = A[inds, None]

File "/home/lpaz/anaconda3/lib/python3.6/site-packages/theano/tensor/var.py", line 572, in __getitem__
return theano.tensor.subtensor.advanced_subtensor(self, *args)

File "/home/lpaz/anaconda3/lib/python3.6/site-packages/theano/gof/op.py", line 674, in __call__
required = thunk()

File "/home/lpaz/anaconda3/lib/python3.6/site-packages/theano/gof/op.py", line 892, in rval
r = p(n, [x[0] for x in i], o)

File "/home/lpaz/anaconda3/lib/python3.6/site-packages/theano/tensor/subtensor.py", line 2190, in perform
rval = inputs[0].__getitem__(inputs[1:])

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.

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
https://github.com/Theano/Theano/issues/6667, or mute the thread
https://github.com/notifications/unsubscribe-auth/AALC-4387ZtMLMYELvj9ckm--si7-2dYks5u37zjgaJpZM4ZNfH4
.

[lucianopaz]

The version of numpy that I'm using is 1.15.4. This does not break theano, it's a regular warning. I just treated it as an exception with this block

with warnings.catch_warnings(record=True) as w:
    # Cause all warnings to always be triggered.
    warnings.simplefilter("error")

to get the full traceback to what was actually raising the warning

[lucianopaz]

Closed by #6703

[AlexAndorra]

Hi @lucianopaz! Is this issue supposed to be resolved now?
Because it looks like the FutureWarning is still here.
Thanks!

[lucianopaz]

@alexandorra, It's fixed on theano's master branch. The thing is that they haven't made a new release of theano to make the fix available to anyone that tries to simply pip install --upgrade theano

[AlexAndorra]

Oooh ok, thanks Luciano! So I guess the fix would have to be made on Theano-PyMC 🤔

[lucianopaz]

@twiecki was planning on cutting a release here

[twiecki]

@lucianopaz Actually I wanted to use this opportunity to cut over to our fork.