[astro] Re: bug in likelihood estimation

[Travis Desell]

In the likelihood estimation, each star calculates:

        //update: check star_prob==0, not prob_sum
        if (star_prob != 0.0) {
            es->prob_sum += log(star_prob)/log(10.0);
        } else {
            es->num_zero++;
            es->prob_sum -= 238.0;
        }

I'm wondering if a slight variation in the parameters could cause a few 
stars to have the star_prob calculation become 0?  This might account 
for the second, more fit line in the graphs.  Since the code is 
subtracting 238 from the prob_sum, this would lead to that point having 
a higher fitness (when it should probably have a lower one).

Could this be causing the bug?  And is there a better way to handle this 
situation?

--Travis


Travis Desell wrote:
> The application we've compiled and made available on BOINC ourselves.
>
> Malik Magdon-Ismail wrote:
>> Hi,
>>
>> What is a stock application?
>>
>> -Malik
>>
>> -------------------------------------------------------------------------- 
>>
>> Have Faith in Probability
>>
>> Malik Magdon-Ismail
>> Associate Professor of Computer Science
>> Rensselaer Polytechnic Institute (RPI)
>>
>>  Office: Rm 312 Lally           mail:   Rm 207, Lally Building,
>>  email:  magdon at cs.rpi.edu              CS Department, RPI,
>>  tel:    518-276-4857                   110 8th Street,
>>  fax:    518-276-4033                   Troy, NY 12180-3590.
>>  web:    http://www.cs.rpi.edu/~magdon
>>  key:    http://www.cs.rpi.edu/~magdon/MalikMagdon-Ismail.PublicKey.asc
>> -------------------------------------------------------------------------- 
>>
>>
>> On Tue, 20 Jan 2009, Travis Desell wrote:
>>
>>> I've been analyzing the data coming in from our stock applications 
>>> and it looks like there's some kind of bug in the likelihood 
>>> calculation.  Here are plots of points along the direction generated 
>>> by the hessian for stripes 79, 82 and 86:
>>>
>>> files with _darwin are points only from os/x clients,  _linux is 
>>> linux clients, and _stock is linux/windows/os x.  As you can see 
>>> they're all having the same issue. It almost seems like two distinct 
>>> lines are being made.
>>>
>>> Does anyone have any idea what might be causing this?  Again, these 
>>> are points only coming from our stock application, so it's not a 
>>> matter of a bad application being out there.  I have a feeling this 
>>> is what's throwing off the hessian calculation.
>>>
>>> -- 
>>> ---------------------------------------------------------------------------------------------------------- 
>>>
>>> Travis Desell
>>> <deselt @ cs.rpi.edu>
>>> 1-518-878-7424
>>> Worldwide Computing Laboratory ( http://wcl.cs.rpi.edu/ )
>>> MilkyWay at Home ( http://milkyway.cs.rpi.edu/ )
>>> Rensselaer Polytechnic Institute, 110 8th Street, Troy NY 12180, USA
>>> ---------------------------------------------------------------------------------------------------------- 
>>>
>>
>>
>

-- 
----------------------------------------------------------------------------------------------------------
Travis Desell
<deselt @ cs.rpi.edu>
1-518-878-7424
Worldwide Computing Laboratory ( http://wcl.cs.rpi.edu/ )
MilkyWay at Home ( http://milkyway.cs.rpi.edu/ )
Rensselaer Polytechnic Institute, 110 8th Street, Troy NY 12180, USA
----------------------------------------------------------------------------------------------------------