Bug #66712 for Statistics-Descriptive: variance calculation is not numerically stable

Fri Mar 18 09:58:35 2011 KUBINA [...] cpan.org - Ticket created

CC:	jeff.kubina [...] gmail.com
Subject:	variance calculation is not numerically stable

Shlomi,

The method to compute the variance is numerically unstable, see http://bit.ly/gNE2pl; version 2 of the module used the method by Knuth and Welford, see http://bit.ly/gY4u5t, to compute the variance but this was removed in the update to version 3.

Would it be possible to move the variance calculation back to the Knuth and Welford method; in that case the variance would never be negative and more accurate?

--
Jeff

Fri Mar 25 03:39:56 2011 SHLOMIF [...] cpan.org - Correspondence added

Hi Jeff, On Fri Mar 18 09:58:35 2011, KUBINA wrote: Show quoted text

> Shlomi, > The method to compute the variance is numerically unstable, see > http://bit.ly/gNE2pl ; version 2 of the module used the method by Knuth > and > Welford, see http://bit.ly/gY4u5t , to compute the variance but this > was removed > in the update to version 3. > > Would it be possible to move the variance calculation back to the > Knuth and > Welford method; in that case the variance would never be negative and > more > accurate? >

If possible, please submit a patch that includes modifications to one of the test scripts for that. I'm not a big expert in the methods of calculating such stuff. Regards, -- Shlomi Fish

Fri Mar 25 03:39:57 2011 The RT System itself - Status changed from 'new' to 'open'

Wed Aug 03 04:15:44 2011 SHLOMIF [...] cpan.org - Correspondence added

RT-Send-CC:

jeff.kubina [...] gmail.com

Hi, Jeff. Please follow my request in the comment. Otherwise, I'll have to close this bug. Regards, -- Shlomi Fish On Fri Mar 18 09:58:35 2011, KUBINA wrote: Show quoted text

> Shlomi, > The method to compute the variance is numerically unstable, see > http://bit.ly/gNE2pl; version 2 of the module used the method by Knuth > and > Welford, see http://bit.ly/gY4u5t, to compute the variance but this > was removed > in the update to version 3. > > Would it be possible to move the variance calculation back to the > Knuth and > Welford method; in that case the variance would never be negative and > more > accurate? > > -- > Jeff

Fri Nov 04 09:08:55 2011 SHLOMIF [...] cpan.org - Correspondence added

On Wed Aug 03 04:15:44 2011, SHLOMIF wrote: Show quoted text

> Hi, Jeff. > > Please follow my request in the comment. Otherwise, I'll have to close > this bug. >

Closing due to lack of responsiveness and inactivity. Please reply (and preferably supply a patch) to re-open. Regards, — Shlomi Fish Show quoted text

> Regards, > > -- Shlomi Fish

Fri Nov 04 09:08:56 2011 SHLOMIF [...] cpan.org - Status changed from 'open' to 'resolved'