Change memory mapped array sampling scheme to match NeuroVault

[tyarkoni]

Currently the memory mapped numpy arrays used in decoding sample ~20k random voxels from the whole brain. NeuroVault resamples images to isotropic 4 mm voxels, which gives ~30k voxels. It would probably speed up processing a bit to use the same scheme in Neurosynth, plus results might be slightly more representative using resampled voxels rather than random ones.

[vsoch]

+1

[vsoch]

priority:awesome

[tyarkoni]

Well, realistically, this might take us from ~0.5s per decoding to ~0.3s, so I'm not sure it's worth upping the priority. But next time I'm mucking around with that part of the code, I'll add this. :)

[vsoch]

The level of awesomeness is not impacted by being low priority! :O)