[pymvpa] hang on poly_detrend

[Brian Murphy]

Hi again,

I looked into the features one by one as you suggested, and it turned 
out that the data itself was crazy (containing numbers that ranged from 
0 to 1x10^218, and then infinity). So the problem must lie further back 
in my processing stream - something that has nothing to do with PyMVPA...

sorry for the false alarm,

best,

Brian
> Date: Thu, 7 Apr 2011 12:07:51 -0400
> From: Yaroslav Halchenko <debian at onerussian.com>
> Subject: Re: [pymvpa] hang on poly_detrend
> To: Development and support of PyMVPA
>         <pkg-exppsy-pymvpa at lists.alioth.debian.org>
> Message-ID: <20110407160751.GR6199 at onerussian.com>
> Content-Type: text/plain; charset=us-ascii
>
> well -- probably the problem in numpy's linalg.lstsq which uses lapack's
> SVD... it is very unlikely that it doesn't converge on reasonable data,
> but I guess could happen... so -- do feature by feature and find on
> which it gets stuck -- share the data -- lets try with different numpy
> versions/lapack builds to see if it is reproducible
>
> On Thu, 07 Apr 2011, Brian Murphy wrote:
>
>   
>> Hi,
>>     
>
>   
>> I've been using poly_detrend on a range of datasets and
>> classification tasks recently, with success. But on my latest script
>> it hangs, and never completes (the step that usually takes ~1s on a
>> dataset of ~500 cases by ~1000 features, didn't complete after
>> several hours). This only happens with a very particular set of
>> parameters (certain datasets, and dimensionalities). Any suggestions
>> why, or how to proceed in tracking the problem down?
>>     
>
>   
>> Brian
>>     
>
>