As I understand it, Olivier is talking about another type of cascade there.
There, the initial and secondary decisions are into the same set of labels.
I've tried building such a generic elimination cascade, but I think the
training regime could go a few different ways.
For a heirarchical label cascade, you'll need an object that takes perhaps
a single estimator, or a different estimator for each hierarchical depth;
it also needs to take the heirarchy as input. You need to train a clone of
the estimator (see sklearn.base.clone and its uses) at each split node in
the hierarchy. Input to fit, and output of predict, should just be the
finest granularity labels, which are first mapped up to the depth 1 types
and successively expanded.
I'm not sure if that's useful. What sort of suggestions were you looking
for?
On 16 January 2014 15:26, Louis Tiao <louist...@gmail.com> wrote:
> Say I have a classification problem that is multiclass and
> characteristically hierarchical, e.g. 'edible', 'nutritious' and
> '~nutritious' - so it can be represented like so
>
> ├── edible│ ├── nutritious│ └── ~nutritious└── ~edible
>
> While one can get reasonable performance with classifiers that support
> multiclass classification or using one-vs-one/all schemes for those that
> don't, it may also be beneficial to separately train classifiers at each
> level and concatenate them so the instances classified as 'edible' can be
> classified as eithernutritious or not.
>
> I would like to use scikit-learn estimators as building blocks and I am
> wondering if I can make thePipeline support this or if I would need to
> write my own estimator that implements the base estimator and possibly
> BaseEnsemble to do this.
>
> It has been mentioned before by @ogrisel on the mailing list
> http://sourceforge.net/mailarchive/message.php?msg_id=31417048 and I'm
> wondering if anyone has insights or suggestions on how to go about doing
> this.
>
> Thanks,
>
> Louis
>
>
> ------------------------------------------------------------------------------
> CenturyLink Cloud: The Leader in Enterprise Cloud Services.
> Learn Why More Businesses Are Choosing CenturyLink Cloud For
> Critical Workloads, Development Environments & Everything In Between.
> Get a Quote or Start a Free Trial Today.
>
> http://pubads.g.doubleclick.net/gampad/clk?id=119420431&iu=/4140/ostg.clktrk
> _______________________________________________
> Scikit-learn-general mailing list
> Scikit-learn-general@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
>
>
------------------------------------------------------------------------------
CenturyLink Cloud: The Leader in Enterprise Cloud Services.
Learn Why More Businesses Are Choosing CenturyLink Cloud For
Critical Workloads, Development Environments & Everything In Between.
Get a Quote or Start a Free Trial Today.
http://pubads.g.doubleclick.net/gampad/clk?id=119420431&iu=/4140/ostg.clktrk
_______________________________________________
Scikit-learn-general mailing list
Scikit-learn-general@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
