Am 06.12.2012 19:44, schrieb Rich Jones:
> Bummer!
>
> Still, this could be something that I'm able to tackle, and if you're 
> willing to help me figure out how to implement it, I'm willing to 
> write all of the docs/tests if necessary, too.
>
> What do you think my plan of attack should be?
Thanks for volunteering to work on this. It would be a great contribution.

As you are interested in the Gaussian version, maybe just start with 
this. I am no expert in the NB business,
but I think it should be quite straight forward.

In the "fit" function, what is done is estimate the mean and variance 
for each feature per class.

So you add a new function called "partial_fit", that does the same, but 
incrementally.
For that you can just also store the incremental statistics of the data 
and update these at each call to
"partial_fit".
For the mean, this is very easy: just keep accumulating and remember how 
many points you had already.
(It might be a bit more stable if you remember the number of points and 
the mean so that the sum doesn't grow
to large).
For the variance, maybe take a look at the wikipedia page:
https://en.wikipedia.org/wiki/Algorithms_for_calculating_variance
I think the "online algorithm" is the thing we want.
Don't hesitate to submit an early pull request, which makes it easier to 
talk about the code
and provides a good platform for discussion.

If you need more input before the start, just ask.

Be sure to read the coding guidelines:
http://scikit-learn.org/dev/developers/index.html#coding-guidelines
and probably also the API docs.

Thanks again,
Andy


------------------------------------------------------------------------------
LogMeIn Rescue: Anywhere, Anytime Remote support for IT. Free Trial
Remotely access PCs and mobile devices and provide instant support
Improve your efficiency, and focus on delivering more value-add services
Discover what IT Professionals Know. Rescue delivers
http://p.sf.net/sfu/logmein_12329d2d
_______________________________________________
Scikit-learn-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general

Reply via email to