That's going a bit overboard here. A stratified K fold which preserves
ordered groups just requires running KFold over samples by target value and
merging across target values.


On Tue, Aug 20, 2013 at 6:06 PM, Lars Buitinck <[email protected]> wrote:

> I think shuffling is a good idea.
>
> 2013/8/20 Olivier Grisel <[email protected]>:
> > Wouldn't it be possible to implement a StratifiedKFolds that preserves
> > the dependency relationship as much as possible?
>
> I have a KFold that preserves group/sequence structure in seqlearn
> [1]. It's not stratified, because the k-fold splitting alone was a
> combinatorial optimization problem (I needed to preserve the sequence
> structure *exactly*); I opted for shuffling + repeats instead. Maybe
> this can serve as inspiration?
>
> [1]
> https://github.com/larsmans/seqlearn/blob/master/seqlearn/evaluation.py#L90
>
> --
> Lars Buitinck
> Scientific programmer, ILPS
> University of Amsterdam
>
>
> ------------------------------------------------------------------------------
> Introducing Performance Central, a new site from SourceForge and
> AppDynamics. Performance Central is your source for news, insights,
> analysis and resources for efficient Application Performance Management.
> Visit us today!
> http://pubads.g.doubleclick.net/gampad/clk?id=48897511&iu=/4140/ostg.clktrk
> _______________________________________________
> Scikit-learn-general mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
>
------------------------------------------------------------------------------
Introducing Performance Central, a new site from SourceForge and 
AppDynamics. Performance Central is your source for news, insights, 
analysis and resources for efficient Application Performance Management. 
Visit us today!
http://pubads.g.doubleclick.net/gampad/clk?id=48897511&iu=/4140/ostg.clktrk
_______________________________________________
Scikit-learn-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general

Reply via email to