predictionio-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Kenneth Chan <>
Subject Re: Help requested: Multiple events and algorithms on one engine
Date Sun, 05 Mar 2017 05:38:41 GMT
whether combining multiple into one or separate them as diff engines really
depends on your use case and design choice.

just like having single program with some logic to handle different things
so they can share some common logic and configured VS individual program
for each one for clean separation and clarity.

On Wed, Mar 1, 2017 at 8:14 AM Pat Ferrel <> wrote:

> No, both engines can even use the same data. Two engines is far simpler to
> construct and manage. You make 2 queries but internally with multiple
> algorithms 2 queries would be made also. The only thing you save is having
> 2 deployed engine processes rather than having them in the same single
> process at the cost if complexity.
> On Feb 28, 2017, at 6:11 PM, Kanak Singh <> wrote:
> This way, I can deploy one engine for one user of my web app, and that can
> take care of all prediction needs of that user.
> If I deploy multiple engines for one user, it will limit the number of
> users I can have, right?
> On Tue, Feb 28, 2017 at 4:49 PM, Kanak Singh <>
> wrote:
> This way, I only have to deploy that one engine, instead of deploying
> multilpe ones.
> On Tue, Feb 28, 2017 at 4:19 PM, Pat Ferrel <> wrote:
> May I ask why you are packaging these as multiple algorithms in a template
> rather than multiple templates with one algo each?
> On Feb 28, 2017, at 3:29 PM, Kanak Singh <> wrote:
> Hi all,
> This is my first email so pardon me for any mistakes in email etiquette.
> I am trying to create a multi purpose engine in which I can include
> multiple algorithms that operate on multiple data sets, and that is what I
> need some help with.
> I have read the documentation on including multiple algorithms and
> handling multiple events. I see that the DataSource and Preparator can
> collect data across different channels/ different event types and pass it
> on as individual RDD[LabeledPoint] fields of the PreparedData object. And I
> can add multiple Algorithm files and have each of them access the
> respective PreparedData field that I choose.
> However, for this I have to decide which algorithms I can offer to the
> client and how many of them, at the time of creating an Engine template
> folder. I am trying to go one step further by making one engine handle an
> unspecified number of data sets (dynamically uploaded by client using POST
> requests) and operate on them using any algorithm (specified perhaps in a
> 'pio train --algorithm' option).
> First of all, is this a good idea?
> If so, what is the best way to do it?
> Any help or leads would be appreciated.
> Best.

View raw message