Inconsistent values of KS-statistics

[vfilimonov]

It seemed that sometimes values of KS statistics, that are stored in _xmin_kstest are different from those that are calculated when plfit.plfit is called with fixed xmin.

For example for the data from https://gist.github.com/vfilimonov/1072e402e922712ad980#file-tst-csv:

y = np.genfromtxt('tst.csv', delimiter=',')
tst_fit = plfit.plfit(y)

def func(xmin):
    ff = plfit.plfit(y, xmin=xmin, quiet=True, silent=True)
    return ff._ks

tst_KS_plfit = [func(xmin) for xmin in tst_fit._xmins]

Values in tst_fit._xmin_kstest and in tst_KS_plfit will be different:

plot(tst_fit._xmins, tst_fit._xmin_kstest, 'r.-')
plot(tst_fit._xmins, tst_KS_plfit, 'g.-')

[keflavich]

That's a small but significant difference... thanks for finding it. I'll try to pin down the problem...

[keflavich]

I can reproduce the error but haven't yet figured out the source of the problem. I thought it might be a failure to use only the unique data in the kstest_ function, but that's not it.

[keflavich]

Part of the problem is that your data are continuous, but the discrete/continuous "guesser" is identifying them as discrete. Pass discrete=False and you'll get a better result... but there is still some underlying inconsistency I haven't resolved.

[vfilimonov]

Adam, thanks.

As a temporary workaround one can manually sweep along xmins - in this case KS-statistics is correct.

On 13 Jun 2014, at 17:15 , Adam Ginsburg [email protected] wrote:

Part of the problem is that your data are continuous, but the discrete/continuous "guesser" is identifying them as discrete. Pass discrete=False and you'll get a better result... but there is still some underlying inconsistency I haven't resolved.

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