You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@predictionio.apache.org by Utkarsh Deep <ut...@gmail.com> on 2018/04/26 06:39:54 UTC

Proposing features for PIO

Hi all,

I am in process of exploring PIO as the framework to deploy ML models to 
production. But to serve our use-case there are few things missing and I 
would love to contribute and would invite anyone who could guide and 
support me in the journey.

List of features that I propose:

1) Implementing Cassandra as the event and metadata store.

2) Adding a monitoring and management  framework for the ML models

     Monitoring: Real time evaluation of the model whenever it produces 
an output given a query. Monitoring can include many things like: health 
monitoring, latencies, business and model metric real time evaluation.

     Management: Versioning of model, keeping track of what went into 
building model, model updation.


Some might already be available, some might not. I would love to hear 
back and contribute back with some guidance.


Regards,

Utkarsh Deep


Re: Proposing features for PIO

Posted by Donald Szeto <do...@apache.org>.
Hi Utkarsh,

These are awesome features to have. I will definitely help provide guidance
and support.

1) I think someone may have C* drivers work-in-progress. If that’s the
case, I invite the person(s) to come forward and join this journey.

2) PIO supports a plug-in architecture in its serving process. For
monitoring, we can start from there. For model management, I believe we
mostly need to expose it. That is already in the metadata.

Regards,
Donald

On Thu, Apr 26, 2018 at 11:03 AM Utkarsh Deep <ut...@gmail.com> wrote:

> Hi all,
>
> I am in process of exploring PIO as the framework to deploy ML models to
> production. But to serve our use-case there are few things missing and I
> would love to contribute and would invite anyone who could guide and
> support me in the journey.
>
> List of features that I propose:
>
> 1) Implementing Cassandra as the event and metadata store.
>
> 2) Adding a monitoring and management  framework for the ML models
>
>      Monitoring: Real time evaluation of the model whenever it produces
> an output given a query. Monitoring can include many things like: health
> monitoring, latencies, business and model metric real time evaluation.
>
>      Management: Versioning of model, keeping track of what went into
> building model, model updation.
>
>
> Some might already be available, some might not. I would love to hear
> back and contribute back with some guidance.
>
>
> Regards,
>
> Utkarsh Deep
>
>