predictionio-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Pat Ferrel <...@occamsmachete.com>
Subject Re: Multiple apps problem
Date Fri, 13 Jan 2017 18:39:41 GMT
You have 2 copies of the UR directory and 2 engine.json files and 2 input event sets? Make
sure there are different “indexName” set for each model in the 2 engine.json files. Otherwise
training of one engine will wipe out the other since they point to the same indexName.

To have 2 completely separate applications with different data you need:
1) 2 different EventServer appNames with different appids, create with `pio app new <your-app-name>`
2) 2 different indexNames for 2 models, set in the 2 engine.json files
3) each engine deployed on a separate port (you have this already)


On Jan 13, 2017, at 12:35 AM, Hossein Bahrami <h.bahrami@live.com> wrote:

Hi Pat,
 
I have a Ubuntu machine which PredictionIo (0.9.7-aml) installed on it,
I have two application both based on Universal recommender template.
I trained both of the apps and deployed them on different ports .
Results of the apps were quite normal but the problem which I figured out later is that one
of the apps was predicting items with score zero so I retrained it then scores become normal
numbers but after this action (retraining) the other app start returning empty results, so
I retrained that one and then the first one returned zero score items again .
 
This is like closed circle and retraining each of apps causes the other app return empty results
or abnormal zero scored results.
 
Regards 
Hossein
 
From: Pat Ferrel <mailto:pat@occamsmachete.com>
Sent: Thursday, January 12, 2017 5:57 PM
To: user@predictionio.incubator.apache.org <mailto:user@predictionio.incubator.apache.org>
Subject: Re: Multiple apps problem
 
This forum is to support Apache PIO, and the latest UR runs on it so I advise you upgrade.
Instructions are here on actionml.com <http://actionml.com/> or here for Apache PIO:
http://predictionio.incubator.apache.org/install/ <http://predictionio.incubator.apache.org/install/>

I don’t understand your explanation, there should be one train per PredictionServer, you
can’t have one PS and two “engines” or configurations. The 2 engines must be on 2 different
ports if deployed separately.

Can you explain your setup and what you are trying to do?


On Jan 12, 2017, at 7:39 AM, Hossein Bahrami <h.bahrami@live.com <mailto:h.bahrami@live.com>>
wrote:

Hi everyone,
 
I have a very strange problem. I have two application running on one predictionio engine,
each time I train one of them the other one’s predictions returning zero score items for
one of them and no items at all for the other.
 
I am using UR algorithm for both of apps
 
pio version : 0.9.7-aml
 
regards


Mime
View raw message