[scikit-learn] Moving average transformer

[Joel Nothman]

I agree that this is best handled with a custom transformer, for the
reasons cited by Jacob, but also because it sounds like this transformer
does not gather statistics from the training data, and so can be
implemented with FunctionTransformer

On 7 Jul 2017 6:10 am, "Jacob Schreiber" <jmschreiber91 at gmail.com> wrote:

Hi Jeremy!

Thanks for your offer to contribute. We're always looking for people to add
good ideas to the package. Time series data can be tricky to handle
appropriately, and so I think we generally try to pass it off to more
specialized packages that focus on that. Andreas may have a more detailed
perspective on this though.

Jacob

On Thu, Jul 6, 2017 at 12:50 PM, jt cunni <jtcunni at gmail.com> wrote:

> First off, I have never contributed to anything before so please have
> patience with me.  I am a data scientist and I have been working with doing
> some feature engineering on one of my datasets.  In my code, I have a
> pipeline of several transformers and an estimator.  I use my pipeline
> and randomizedsearchcv to tune my hyper-parameters and my transformer
> settings. Pretty standard stuff.  One thing I was doing was creating a
> feature that was a moving average of another feature. In a basic example,
> imagine I want to predict if a team is going to win a baseball game.  I
> create a feature that is the moving average of the last N games of runs
> scored per game (this is the window size of the moving average).  Not
> knowing what the best window size for the moving average, I created a
> custom transformer that could be put in a pipeline to find the window size
> that provides the most lift.  Is there any interest for this type of
> contribution? If so, what unittests or anything else do I need to provide?
>
>
>
> Thanks,
>
> Jeremy
>
> _______________________________________________
> scikit-learn mailing list
> scikit-learn at python.org
> https://mail.python.org/mailman/listinfo/scikit-learn
>
>

_______________________________________________
scikit-learn mailing list
scikit-learn at python.org
https://mail.python.org/mailman/listinfo/scikit-learn
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/scikit-learn/attachments/20170707/01cd9999/attachment.html>