Open Stack

> Not sure what you mean by serious.
>
> Maybe you could have a look at Meteos[1]. It is a young project but surely
> focuses on machine learning.
>
> [1]: https://wiki.openstack.org/wiki/Meteos
Another avenue is to use Storlets for either the learn or prediction  
phase where the data resides in Swift.
We are currently adding IPython integration [1] that makes it very  
easy to deploy and invoke Storlets from IPython (a data scientists  
beloved tool :-), plus [2] is an initial working towards leveraging  
Storlets for machine learning.

In few more words: Storlets [3] allow to run a serverless computation  
inside Swift nodes, where the computation is done inside a Docker  
container. This basically means that you can write a piece of code (in  
either Python or Java) upload that code to Swift (as if it was a data  
object) and then invoke the uploaded code (called storlet) on your  
data (much like AWS Lambda). The nice thing is that the Docker image  
where the storlet is executed can be tailored by the admin, and as to  
make sure it has, e.g. scikit-learn installed. With such a Docker  
image you can write a storlet that would use the sickit-learn  
algorithms on swift objects.


[1] https://review.openstack.org/#/c/416089/
[2] https://github.com/eranr/mlstorlets
[3] http://storlets.readthedocs.io/en/latest/