[SciPy-Dev] weird results in frozen distribution fit
josef.pktd at gmail.com
josef.pktd at gmail.com
Thu Oct 20 09:01:47 EDT 2011
On Thu, Oct 20, 2011 at 7:25 AM, K.-Michael Aye <kmichael.aye at gmail.com> wrote:
> So, Josef, are you saying that we should not use frozen fits for the
> moment? I consider you one of the authorities on these matters, me as a
> fitting beginner am mostly lost in the choices I have.
> Do we have alternatives/work-arounds?
Yes, that's what I'm saying. The results are wrong in all examples
that I tried for fitting with some frozen parameters with the scipy
0.9 release.
Per Brodtkorb provided the bugfix in the ticket. It's just changing
the indexing in a few lines.
If you want, you could just change the few lines in your installed scipy.
There is no workaround outside changing the source that I can think of.
(as a reminder: testing this part was not my responsibility.)
Josef
>
> Michael
>
> On 2011-10-10 22:22:27 +0000, josef.pktd at gmail.com said:
>
>> Given a report on the scipy-user mailing list
>> http://projects.scipy.org/scipy/ticket/1536 I started for the first
>> time to look at some examples with the frozen fit introduced in scipy
>> 0.9. I just picked randomly some distributions that came to mind,
>> except for the reported lognorm.
>>
>> some cases look ok, maybe. Some cases look "weird"
>>
>> Josef
>>
>> lognorm true
>> 0.25 0.0 20.0
>> estimated, floc=0, loc=0
>> [ 2.1221 0. 2.4403] [ 0.2303 -1.8759 21.9986]
>> [ 2.1454 0. 2.3763] [ 0.2446 0.0305 19.9945]
>> [ 2.1274 0. 2.414 ] [ 0.2469 0.0296 19.9415]
>> [ 2.1362 0. 2.3897] [ 0.2525 0.4579 19.4758]
>> [ 2.1334 0. 2.4104] [ 0.2484 0.0298 20.0291]
>> [ 2.1033 0. 2.475 ] [ 2.5335e-01 1.0858e-02 1.9959e+01]
>> [ 2.1266 0. 2.414 ] [ 0.2662 1.3626 18.5458]
>> [ 2.1316 0. 2.4176] [ 0.2491 0.0299 20.0509]
>> [ 2.0897 0. 2.5211] [ 0.276 1.4881 18.5202]
>> [ 2.1382 0. 2.3738] [ 0.2465 0.0301 19.8241]
>> gamma true
>> 2 20 10.0
>> estimated, floc=0, loc=0
>> [ 2.0945 20. 9.6182] [ 1.9864 20.0832 10.2119] [ 2.0317
>> 20. 10. ]
>> [ 2.0084 20. 10.101 ] [ 1.9364 20.1738 10.41 ] [ 2.0241
>> 20. 10. ]
>> [ 2.1359 20. 9.1555] [ 2.0353 19.8413 9.8974] [ 1.9937
>> 20. 10. ]
>> [ 2.2296 20. 8.6785] [ 2.152 19.8834 9.1795] [ 1.9948
>> 20. 10. ]
>> [ 1.8822 20. 10.8936] [ 2.0389 19.9836 9.8365] [ 2.0104
>> 20. 10. ]
>> [ 1.8304 20. 11.3802] [ 2.0341 20.169 9.7599] [ 2.0214
>> 20. 10. ]
>> [ 1.972 20. 9.7958] [ 1.9107 20.0777 10.1223] [ 1.9409
>> 20. 10. ]
>> [ 1.6211 20. 13.3028] [ 1.9636 20.1903 10.1156] [ 2.0103
>> 20. 10. ]
>> [ 2.0413 20. 9.8165] [ 2.0238 19.9456 9.9794] [ 2.0121
>> 20. 10. ]
>> [ 2.0087 20. 10.1362] [ 2.0128 20.1212 9.9924] [ 2.0299
>> 20. 10. ]
>> normal true
>> 0.0 2.0
>> estimated, floc=0, loc=0, fscale=2
>> [ 0. 2.0024] [-0.0417 2.0024] [-0.0417 2. ]
>> [ 0. 2.0003] [-0.0879 2.0003] [-0.0879 2. ]
>> [ 0. 1.9466] [-0.0133 1.9466] [-0.0133 2. ]
>> [ 0. 1.9726] [-0.036 1.9726] [-0.0359 2. ]
>> [ 0. 2.0007] [-0.0194 2.0007] [-0.0195 2. ]
>> [ 0. 1.9565] [ 0.0338 1.9565] [ 0.0337 2. ]
>> [ 0. 1.9674] [ 0.0288 1.9674] [ 0.0289 2. ]
>> [ 0. 1.9962] [ 0.0051 1.9962] [ 0.0051 2. ]
>> [ 0. 1.8901] [ 0.0179 1.8901] [ 0.0179 2. ]
>> [ 0. 2.0017] [ 0.0307 2.0017] [ 0.0307 2. ]
>> chi2 true
>> 10 0.0 2.0
>> estimated, floc=0, loc=0
>> [ 3.8459 0. 6.1457] [ 10.9616 -0.6938 1.8646]
>> [ 3.8314 0. 6.3129] [ 10.4894 -0.295 1.9516]
>> [ 3.7449 0. 6.4353] [ 9.7304 -0.0364 2.0654]
>> [ 3.8094 0. 6.2268] [ 10.4314 -0.3388 1.9283]
>> [ 3.69 0. 6.5633] [ 8.4951 1.1035 2.2316]
>> [ 3.6475 0. 6.6415] [ 8.3315 0.8241 2.3152]
>> [ 3.7288 0. 6.4983] [ 9.4114 0.2128 2.1185]
>> [ 3.8005 0. 6.2852] [ 10.0038 -0.0154 1.9917]
>> [ 3.7384 0. 6.4327] [ 8.4504 1.1931 2.224 ]
>> [ 3.9063 0. 6.0929] [ 10.3413 0.2105 1.902 ]
>>>>>
>>
>> _______________________________________________
>> SciPy-Dev mailing list
>> SciPy-Dev at scipy.org
>> http://mail.scipy.org/mailman/listinfo/scipy-dev
>
>
>
> _______________________________________________
> SciPy-Dev mailing list
> SciPy-Dev at scipy.org
> http://mail.scipy.org/mailman/listinfo/scipy-dev
>
More information about the SciPy-Dev
mailing list