Gael Varoquaux

I would love this. But I believe that last time I looked, I found that it was actually fairly technical.

You should give us some code to reproduce the problem. I suspect that it is just that you are trying to process data that is too big for your memory.

I use them when teaching: they avoid installation mess in the room. On Aug 10, 2022, 17:28, at 17:28, Andrei Foldes ***@***.***> wrote: >I do use it when quickly wanting...

It can be obtained by using a masked array. Probably combining the numpy masked array tools (https://docs.scipy.org/doc/numpy-1.15.1/reference/maskedarray.html) with nilearn.image.math_img is the most elegant way of doing it. I can see...

Examples of repo where this work: scikit-learn, MNE-python

@agramfort : thanks, I'll let @NicolasGensollen ping you if he gets to this.

Weighted average of the neighbors, with the weight being function of the distance between the two nodes (approximates a spatial kernel on the mesh). More smoothing would be achieved by...

Given that nilearn is used by many different populations, with different expectations, I think that adding such an option would indeed be a good thing. We would welcome a pull...

Given how nilearn has spread and is popular, I am +1 on integrating this today. As a pointer, it probably should be done by using ax.invert_axis in places such as...

AFAICT, PoSCE now exposed a fit_transform which returns the tangent-space parametrization. The drawback of this is that it confuses users between the covariance estimator and the covariance parametrization. Here is...