Re: [bug #7641] Infinite Chi2 warning -- November 07, 2006 - 09:01

Alexandar Hansen wrote:

> On 11/6/06, *Alexandar Hansen* <viochemist@xxxxxxxxx 
> <mailto:viochemist@xxxxxxxxx>> wrote:
>
>     Whoa!  I thought I had changed my 'runs' lines to not choose
>     anything higher than tm3.  I tried what you said and reduced them
>     models to tm0-tm3 and it still failed.  Through some process of
>     elimination, the one that spits out the error is only tm3 (tm, S2,
>     and Rex).  tm2 finishes without any error. 
>
>
>
> I received another chi-squared = infinity value when running the 
> prolate, oblate, and ellipsoid models.  I used only tm0, tm1, and tm2 
> for the local_tm selection and then I used m0-m5 for the diffusion 
> model selection.  The chi-square = infinity error only occured on the 
> second round of the minimization for prolate (ie full_analysis.py 
> finished twice but failed the 3rd time for prolate) and on the first 
> round for oblate and ellipsoid.  The sphere model finished without an 
> error.


Dear Alex Ed and Chris

This is the regression which I  found via my first regression test. I 
will try and backport the fix  to the 1.2 and 1.3 branches in the next 
couple of hours. Unfortunatley I can't do it immediately as the machine 
I use for external ssh access here is not responding ;-) The hotfix is 
to find the function isInf in float.py and replace it with

def isInf(obj):

   ''' check to see if a python float is an infinity

       the check returns true for either positive or negative infinities

       obj -- float object to check for infinity

       returns -- True if object is an infinity

       throws -- throws a TypeError if obj isn't a python float

   '''

   # bad result for code checking

   result = None

   # check to see if exponent is all ones (excluding sign bit)

   # if exponent is not all ones this can't be a Inf

   if not isExpAllOnes(obj):

       result =  False

   else:

       # get the mantissa as a byte array properly masked

       manBytes = getMantissaBytes(obj)

       for byte in manBytes:

           #check if any of the unmasked mantissa bytes are zero

           # to be a NaN the mantissa must be zero

           if byte > 0:

               result = False

               break

           result = True

   return result


regards
gary

--
-------------------------------------------------------------------
Dr Gary Thompson
Astbury Centre for Structural Molecular Biology,
University of Leeds, Astbury Building,
Leeds, LS2 9JT, West-Yorkshire, UK             Tel. +44-113-3433024
email: garyt@xxxxxxxxxxxxxxx                   Fax  +44-113-2331407
-------------------------------------------------------------------