How to implement unsupervised or reinforcement algorithm in new org.apache.spark.ml

[Egor Pahomov <pa...@gmail.com>]

Hi, I'm redoing my PR <https://github.com/apache/spark/pull/2731> about
genetic algorithm in new org.apache.spark.ml architecture. Do we have
already some code about handling unsupervised or reinforcement algorithm in
new architecture? If no do we have some tickets on this matter? If no do we
have understanding when it would be doing, and how?

-- 



*Sincerely yoursEgor PakhomovScala Developer, Yandex*

Re: How to implement unsupervised or reinforcement algorithm in new org.apache.spark.ml

[Joseph Bradley <jo...@databricks.com>]

Hi,

There are no examples currently.  For unsupervised learning, I think the
pattern is straightforward.  It would follow the pattern from supervised
learning, but without the label input column and with a model having a
different transform() behavior.

Reinforcement learning might take a bit more design since I haven't seen
work on it so far.  I'd recommend making a Discussion JIRA to post a set of
requirements and get feedback on a design.  Reinforcement learning would be
great to have in MLlib.

Joseph

On Mon, Mar 9, 2015 at 5:21 AM, Egor Pahomov <pa...@gmail.com> wrote:

> Hi, I'm redoing my PR <https://github.com/apache/spark/pull/2731> about
> genetic algorithm in new org.apache.spark.ml architecture. Do we have
> already some code about handling unsupervised or reinforcement algorithm in
> new architecture? If no do we have some tickets on this matter? If no do we
> have understanding when it would be doing, and how?
>
> --
>
>
>
> *Sincerely yoursEgor PakhomovScala Developer, Yandex*
>