lisa-lab/pylearn2

Fix RemoveMean with axis=1

Opened this issue · 5 comments

lamblin commented

Reported first in gh-1260.

Test case to be added:

import pylearn2.datasets.preprocessing
import pylearn2.datasets.cifar10

ds = pylearn2.datasets.cifar10.CIFAR10(which_set="train")
pp = pylearn2.datasets.preprocessing.RemoveMean(axis=1)
ds.apply_preprocessor(pp, can_fit=True)

Fix: in https://github.com/lisa-lab/pylearn2/blob/master/pylearn2/datasets/preprocessing.py#L668, replace self._mean = ... by self._mean = X.mean(axis=self._axis, keepdims=True). This means that self._mean will be a column or a row, depending on the value of self._axis, instead of a vector, so it can be applied on the correct dimension of X.

akshayah3 commented

@lamblin Is this issue still up for grabs? And where should we add the test case you pointed above?

lamblin commented

@akshayah3 Sure, thanks for volunteering!
The test can be added in pylearn2/datasets/tests/test_preprocessing.py. It would be better to use a different class than CIFAR10 in the test itself, for instance pylearn2.testing.datasets.random_dense_design_matrix.

alisye commented

@lamblin can I work on this?

Deleted user commented

can i work on this?

nouiz commented