TypeError with fits using mpmath about powerlaw

[jeffalstott]

What version of mpmath do you have? Using 0.17, I get this:

In [26]: gammainc = vectorize(gammainc)

In [27]: x = arange(6,51)

In [28]: gammainc(1-2.45, .033*x)
Out[28]:
array([mpf('4.7254650873107051'), mpf('3.5574132536305769'),
mpf('2.7632640267912771'), mpf('2.1986605090749198'),
mpf('1.7831274678121478'), mpf('1.4687544156762053'),
mpf('1.2254992062468544'), mpf('1.0337152450792615'),
mpf('0.88009970283005678'), mpf('0.75537780707828139'),
mpf('0.65291978565420106'), mpf('0.56788337972645819'),
mpf('0.49666484074349937'), mpf('0.43653747013455607'),
mpf('0.38540772512153532'), mpf('0.34164703962281801'),
mpf('0.30397358868852165'), mpf('0.27136770907936236'),
mpf('0.24301044045076509'), mpf('0.21823822803465681'),
mpf('0.19650910210920391'), mpf('0.17737712575477133'),
mpf('0.16047287852110079'), mpf('0.14548840016396494'),
mpf('0.13216546714630714'), mpf('0.12028638548409158'),
mpf('0.10966670188705738'), mpf('0.10014939043977027'),
mpf('0.09160018378579142'), mpf('0.083903799013133412'),
mpf('0.076960868105193828'), mpf('0.070685427059029596'),
mpf('0.065002850862411296'), mpf('0.05984814647835663'),
mpf('0.055164534957336864'), mpf('0.050902268325669228'),
mpf('0.047017638102363639'), mpf('0.043472140994015217'),
mpf('0.040231774111453643'), mpf('0.037266437390975397'),
mpf('0.034549425122539223'), mpf('0.032056991840105457'),
mpf('0.029767980507127756'), mpf('0.027663503079465909'),
mpf('0.025726665261124014')], dtype=object)

How confident are you that those are the alpha and Lambda values that get to gammainc? If that really is the case, it may be an mpf issue.

[deeplycloudy]

I saw the issue with mpmath 0.17 from PyPI, as well as on yesterday's github master. I reproduced it again this evening using your pure mpmath/numpy code above.

Suspecting numpy 1.6.1, I moved over to numpy 1.7.1, and the problem went away (on github master) so I suspect the problem was something subtle in numpy.

Thanks for your help, and for the library. Lots of power laws and gamma distributions in atmospheric science, and this is really easy to use and well-supported by literature.

I'm closing the issue since this isn't a problem with powerlaw itself.

[jeffalstott]

Disconcerting, but ok. Were you having this issue with one particular dataset, or virtually everything?

[deeplycloudy]

This happened again last night after I thought it was fixed. It took place when passing a float32 array into vectorized gammainc. Casting to float64 fixed it.

On Oct 25, 2013, at 6:19, jeffalstott [email protected] wrote:

Disconcerting, but ok. Were you having this issue with one particular dataset, or virtually everything?

—
Reply to this email directly or view it on GitHub.

[jeffalstott]

Interesting. Is this just with that particular dataset?

On Fri, Oct 25, 2013 at 1:06 PM, deeplycloudy [email protected]:

This happened again last night after I thought it was fixed. It took place
when passing a float32 array into vectorized gammainc. Casting to float64
fixed it.

On Oct 25, 2013, at 6:19, jeffalstott [email protected] wrote:

Disconcerting, but ok. Were you having this issue with one particular
dataset, or virtually everything?

—
Reply to this email directly or view it on GitHub.

—
Reply to this email directly or view it on GitHubhttps://github.com//issues/6#issuecomment-27085033
.

[deeplycloudy]

It fails with a relatively generic test (positive or negative z)

In [4]: gammainc=np.vectorize(mpmath.gammainc)

In [5]: z = -1.9

In [6]: x=np.arange(.3, 50, .1)

In [7]: x.dtype
Out[7]: dtype('float64')

In [8]: gammainc(z,x)
Out[8]: array([mpf('3.0704147576940444'), mpf('1.5206831120098647'), ...

In [9]: gammainc(z,x.astype('f4'))
TypeError: cannot create mpf from 0.30000001

On Fri, Oct 25, 2013 at 7:11 AM, jeffalstott [email protected]:

Interesting. Is this just with that particular dataset?

On Fri, Oct 25, 2013 at 1:06 PM, deeplycloudy [email protected]:

This happened again last night after I thought it was fixed. It took
place
when passing a float32 array into vectorized gammainc. Casting to
float64
fixed it.

On Oct 25, 2013, at 6:19, jeffalstott [email protected]
wrote:

Disconcerting, but ok. Were you having this issue with one particular
dataset, or virtually everything?

—
Reply to this email directly or view it on GitHub.

—
Reply to this email directly or view it on GitHub<
https://github.com/jeffalstott/powerlaw/issues/6#issuecomment-27085033>
.

—
Reply to this email directly or view it on GitHubhttps://github.com//issues/6#issuecomment-27085436
.

[jeffalstott]

Thanks for the info. You've successfully narrowed it down to being an mpmath issue. If you open an issue with them can you link it here?

https://github.com/fredrik-johansson/mpmath/
I don't know if they prefer there, their mailing list, or the Google Code site.