Multi-column examples

[Aditi Nabar]

Hi! Are there any examples of how you would do feature engineering on a data set like the Titanic survivors in KeystoneML? All the examples I've seen on the website so far, are for a single column of text or image data. I'd like to see how KeystoneML would work with multiple columns of different types, where you might want to engineer each column differently. Thanks for any pointers!

Aditi

[Evan R. Sparks]

Hi Aditi,

This is a great question. One approach is to break down each processing step (e.g. each featurization step) into a separate "branch" of a processing pipeline and then use `Pipeline.gather` to combine the features before ultimately fitting a model.

I put together a working (well, compiling, at least!) example here - relevant lines highlighted: https://github.com/etrain/keystone/blob/titanicexample/src/main/scala/pipelines/tabular/TitanicSurvivors.scala#L39-L64

Does this accomplish the kind of thing you're looking for?

- Evan

On Fri, Jul 22, 2016 at 8:54 PM, Aditi Nabar <aditi...@gmail.com> wrote:

Hi! Are there any examples of how you would do feature engineering on a data set like the Titanic survivors in KeystoneML? All the examples I've seen on the website so far, are for a single column of text or image data. I'd like to see how KeystoneML would work with multiple columns of different types, where you might want to engineer each column differently. Thanks for any pointers!

Aditi

--
You received this message because you are subscribed to the Google Groups "KeystoneML Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to keystoneml-use...@googlegroups.com.
To post to this group, send email to keystone...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/keystoneml-users/c294483a-577e-4f2d-9dd4-c1befd95f99a%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Aditi Nabar]

Thanks, Evan! That's very helpful. 

Aditi