[astro] Re: BUGG??
Travis Desell
deselt at cs.rpi.edu
Mon Jun 18 19:34:35 EDT 2007
Thats strange, maybe there is a bug in the line search... I'll look into
that. I'm pretty sure i'm calculating
the gradient correctly. Maybe you could send me the matlab code? RPI
mail strips attachments from
the mailing list so you'd have to email it directly to me.
--Travis
Malik Magdon-Ismail wrote:
> Hi,
>
> I am in my office now if you want to discuss what is going on in my code.
>
> -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.
> http://www.cs.rpi.edu/~magdon
> --------------------------------------------------------------------------
>
> On Mon, 18 Jun 2007, Malik Magdon-Ismail wrote:
>
>
>> Hi,
>> I am attaching my results for conjugate gradient. Notice (in accord with
>> the theorem I told you about) after 20 iterations, the optimization is
>> basically done. The function value attained is 10^(-17)!
>>
>> I have attached the values I used for a and x0 (the start value).
>> stats.dat gives the optimization path. The first column is iteration
>> number and the second is function value, the third is gradient norm and
>> the 4th is #function evals for that iteration. Ignore the last 2 columns
>> for now.
>>
>> The zipped directory contains my matlab implementation. I have the C
>> version but I wrote it so long ago that it uses outdated compilers and
>> hard wired paths so it would take longer to get that up. Please compare my
>> code with yours to find the discrepancy.
>>
>> -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.
>> http://www.cs.rpi.edu/~magdon
>> --------------------------------------------------------------------------
>>
>> -- Binary/unsupported file stripped by Ecartis --
>> -- Type: APPLICATION/postscript
>> -- File: funcplot.eps
>>
>>
>> -- Binary/unsupported file stripped by Ecartis --
>> -- Type: APPLICATION/postscript
>> -- File: iterplot.eps
>>
>>
>> -- Attached file included as plaintext by Ecartis --
>> -- File: a.dat
>>
>> 7.9482108e+00
>> 9.5684345e+00
>> 5.2259035e+00
>> 8.8014221e+00
>> 1.7295614e+00
>> 9.7974690e+00
>> 2.7144726e+00
>> 2.5232935e+00
>> 8.7574190e+00
>> 7.3730599e+00
>> 1.3651874e+00
>> 1.1756687e-01
>> 8.9389797e+00
>> 1.9913807e+00
>> 2.9872301e+00
>> 6.6144258e+00
>> 2.8440859e+00
>> 4.6922429e+00
>> 6.4781123e-01
>> 9.8833494e+00
>>
>> -- Attached file included as plaintext by Ecartis --
>> -- File: x0.dat
>>
>> 6.2323385e-01
>> 7.9904862e-01
>> 9.4088994e-01
>> -9.9209174e-01
>> 2.1203515e-01
>> 2.3788207e-01
>> -1.0077634e+00
>> -7.4204475e-01
>> 1.0822950e+00
>> -1.3149970e-01
>> 3.8988049e-01
>> 8.7987107e-02
>> -6.3546523e-01
>> -5.5957330e-01
>> 4.4365349e-01
>> -9.4990380e-01
>> 7.8118162e-01
>> 5.6896065e-01
>> -8.2171429e-01
>> -2.6560685e-01
>>
>> -- Attached file included as plaintext by Ecartis --
>> -- File: stats.dat
>>
>> 1.0000000e+00 5.2425449e+00 3.9243295e+01 1.5000000e+01 1.0000000e-01 1.1773000e-02
>> 2.0000000e+00 1.1422878e+00 8.6765273e+00 1.2000000e+01 1.0000000e-01 8.4690000e-03
>> 3.0000000e+00 3.2193414e-01 3.9091104e+00 1.2000000e+01 1.0000000e-01 3.0930000e-03
>> 4.0000000e+00 1.5989415e-01 1.8297402e+00 1.2000000e+01 1.0000000e-01 3.0980000e-03
>> 5.0000000e+00 8.0912216e-02 9.1901998e-01 1.5000000e+01 1.0000000e-01 3.1190000e-03
>> 6.0000000e+00 2.2746551e-02 8.3362898e-01 1.8000000e+01 1.0000000e-01 3.3830000e-03
>> 7.0000000e+00 1.0389178e-02 4.7310459e-01 1.0000000e+01 1.0000000e-01 4.4770000e-03
>> 8.0000000e+00 2.4818736e-03 3.1626082e-01 1.1000000e+01 1.0000000e-01 2.9920000e-03
>> 9.0000000e+00 9.6525194e-04 1.8502713e-01 1.3000000e+01 1.0000000e-01 3.2770000e-03
>> 1.0000000e+01 4.9940846e-04 7.9555582e-02 1.5000000e+01 1.0000000e-01 1.9650000e-03
>> 1.1000000e+01 3.5056829e-04 4.2293092e-02 1.7000000e+01 1.0000000e-01 3.4550000e-03
>> 1.2000000e+01 3.1706569e-04 2.4632972e-02 1.9000000e+01 1.0000000e-01 4.4420000e-03
>> 1.3000000e+01 2.4816038e-04 1.9296104e-02 1.8000000e+01 1.0000000e-01 3.2560000e-03
>> 1.4000000e+01 1.8129217e-04 2.3905342e-02 1.7000000e+01 1.0000000e-01 3.1940000e-03
>> 1.5000000e+01 2.8580293e-05 3.4871917e-02 1.5000000e+01 1.0000000e-01 3.1370000e-03
>> 1.6000000e+01 5.9731456e-06 2.4108632e-02 1.8000000e+01 1.0000000e-01 3.6450000e-03
>> 1.7000000e+01 3.1553418e-07 7.5515993e-03 2.1000000e+01 1.0000000e-01 3.9350000e-03
>> 1.8000000e+01 9.6678422e-08 1.8111476e-03 2.6000000e+01 1.0000000e-01 3.5400000e-03
>> 1.9000000e+01 7.2590362e-10 1.3768240e-03 2.7000000e+01 1.0000000e-01 3.6460000e-03
>> 2.0000000e+01 2.3907847e-17 8.9659080e-05 3.5000000e+01 1.0000000e-01 3.6290000e-03
>>
>> -- Binary/unsupported file stripped by Ecartis --
>> -- Type: APPLICATION/x-zip-compressed
>> -- File: hessian_conj_grad.zip
>>
>>
>>
>>
>>
>
>
>
--
----------------------------------------------------------------------------------------------------------
Travis Desell
<deselt @ cs.rpi.edu>
1-518-878-7424
Worldwide Computing Laboratory ( http://wcl.cs.rpi.edu/ )
Rensselaer Polytechnic Institute, 110 8th Street, Troy NY 12180, USA
----------------------------------------------------------------------------------------------------------
More information about the astro
mailing list