Is support for memory maps discontinued?

[Eleftherios Garyfallidis]

Hi all,

In the past we could use memory maps with ease directly from Nibabel. My understanding is that this is not any more supported.
Is there an alternative approach?

[Chris Markiewicz]

I assume we're discussing spatial images? Here's the breakdown:

• img.dataobj.get_unscaled() will return a memmap if the data are uncompressed, and will not apply scaling
• np.asanyarray(img.dataobj) will return a memmap if the data are uncompressed and there are no scale-factors
  • img.dataobj[:] would also work, but is less explicit
• np.asanyarray(img.dataobj, target_dtype) will return a memmap if the data are uncompressed, with no scaling, in that dtype
• img.get_fdata(dtype=target_dtype) will return a memmap if the data are uncompressed, with no scaling, in that dtype
  • This will also store a cache of the array on the img object.
• img.get_fdata() will return a memmap if the data are uncompressed, float64, with no scale factors

[Eleftherios Garyfallidis]

Yes. Spatial images. Thank you for the quick response @effigies.