[Pkg-exppsy-pynifti] [Nipy-devel] Data packaging, a proposal for feedback

[Gael Varoquaux]

On Thu, Jul 23, 2009 at 11:09:01PM -0700, Matthew Brett wrote:
> Hi everyone,

> Gael and Fernando and I have been struggling with data packaging, and
> getting some useful feedback from our Debian friends (hello Michael,
> hello Yarick).

> We've thought for a while about this, and come up with the attached
> proposal - formatted version here:

> https://cirl.berkeley.edu/mb312/data_docs/

In general it looks like a good programmer-level API. A few questions on
how we will set things up so that they work out of the box for users
(your document talks more about the way things are set up, then the way
they are exposed to users):

    - We should provide a default template and atlas environment, where
      the default (or 'core') nipy data set gets installed.
    - For this default data set, we should be able to retrieve files
      really easily: 
	    - no repository instantiation
	    - sensible error management (of course this needs to be
	      defined based on experience)
	    - easy programmatic browsing of the available data 
	      (for example a method that returns a nested set of 
	      dictionnaries, or something. I attached an example).
	    - easy interactive browsing of the files (also in example
	      attached)
    - I found through my interactions with users (at least French ones)
      that the notion of repository was confusing. We should not have to
      expose it to end users. However, this does not mean that the model
      needs to change in any way, we could have the following API:
      
	from nipy import load_img, templates # Notice no 'repo'
	mni152t1 = load_img(templates.get('templates', '1mm', 'T1'))

Gaël