[Nipy-devel] Testing failing in formulas (was: Re: nipy without pynifti)

Matthew Brett matthew.brett@gmail....
Mon Nov 30 14:28:49 CST 2009


Hi,

>> For neurospin this is an important piece of code. Indeed, what users do
>> most often is run GLM analysis. The two important pieces of code are the
>> model estimation, and the model specification. To code the specification,
>> that is the design matrix, in a clean and readable way, is actually
>> non-trivial, but I am sure you are perfectly aware of this. So yes, this
>> is a core functionality for most users.
>>
>
> Yes, it will be important for users. I guess I meant that it might better be
> thought of as part of the core of the statsmodels scikit with some
> particular adaptations for nipy users.

I completely agree with Gael in his characterization of code as
teaching.  I think what we can do is be dumb students, and try to
grasp the underlying ideas and how you are thinking of them, in order
to come back to you with the points that we are lost, and, as we begin
to understand, offer suggestions about the way that abstractions might
be expressed or hidden, to help us understand.   I offer myself as
such as dumb student who's main claim to usefulness is that I've often
tried to explain basic statistics to students - but you well know my
limitations as we get closer to reality (I mean the mathematics).

>> All that means is that we should "own" this piece of code a bit more, in
>> other words, understand it better.
>
> Also a good idea. How can I help you (whoever "we" refers to above)
> understand it better (besides adding more comments / docstrings)?

We - in practice, and for now - means those of us looking at the code
with the intention of working on it, that will be me, Gael, and
Fernando, if he has time this week.

Comments and docstrings would be a great help...

Do you have some day this week we could work together on the code,
even if remotely?  For the release I mean?  It would be an enormous
help if so.

See you,

Matthew



More information about the Nipy-devel mailing list