2012/6/27 Florian Hönig <[email protected]>:
> Dear list,
>
> analogously to sklearn.preprocessing.scale and sklearn.preprocessing.Scaler,
> I would like to add something for scaling the individual features to
> the interval [0;1].
>
> I have encountered a number of datasets where mean/variance scaling didn't 
> help
> much for SVM/SVR, while scaling to [0;1] worked miraculously.
>
> Would that be appreciated, and if yes, how should I proceed?

Please submit a new pull request for this:

http://scikit-learn.org/dev/developers/index.html#contributing-code

Don't forget to include some tests and it would be great if you could
write an example that compares the outcome of a SVC or SVR model with
different scaling (using one of the datasets available by default).

> A separate function interval_scale and a separate class IntervalScaler add
> redundant code, but I presume that this would preferred to
> generalizing the present scale/Scaler, right?

I think so, otherwise the constructor parameters will be too complicated to set.

> Btw, I think there is a bug in preprocessing.Scaler.fit. As no
> transformation should be done
> at this point, line 207 in preprocessing.py should be removed:
>
> inplace_csr_column_scale(X, 1 / self.std_)

Please create a dedicated pull request for this with a new test that
would failed without you proposed change.

-- 
Olivier
http://twitter.com/ogrisel - http://github.com/ogrisel

------------------------------------------------------------------------------
Live Security Virtual Conference
Exclusive live event will cover all the ways today's security and 
threat landscape has changed and how IT managers can respond. Discussions 
will include endpoint security, mobile security and the latest in malware 
threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/
_______________________________________________
Scikit-learn-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general

Reply via email to