predictionio-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Pat Ferrel <>
Subject Re: pio train error java.lang.NegativeArraySizeException
Date Wed, 29 Mar 2017 19:46:50 GMT
Do you know an item is in NY? or are you trying to find what is preferred by New Yorkers?

If the latter send events (user-id, location-preference, location-id) as secondary events.
The location-id should be chosen to include some people but not too many so there is variation
in what New Yorkers prefer vs Las Vegas denizens. I’d use postal codes (zip codes) or phone
area codes for coarser granularity. The benefit of using these code-systems is that they are
never dups like 2 cities with the same name. Also lat-lon doesn’t work since not 2 people
can existing in such a fine grained location.  

Once you determine the location encoding scheme you’ll use, determine how many possible
location-ids you’ll have. This should affect the param maxCorrelatorsPerItem described here: <> Set it to a
number much smaller than the total number of location-ids you have. For instance if you were
using user gender you'd set this to 1 (actually this is an extreme and would use another method
but serves to illustrate)

BTW UR support is here:!forum/actionml-user <!forum/actionml-user>

On Mar 29, 2017, at 12:31 PM, InfoQuest Solutions <> wrote:


I need to store event data based on user location like user = 1 coming from New York viewed
an item, then user = 1 coming from New York bought item. I want to recommend items to other
users are viewing in New York itself.

So please help me out with indicator params in engine.json and how to efficiently set event,user
and item data.


So how shall i proceed with set event or i need to put every item against every location and

> On 28-Mar-2017, at 11:35 PM, Haddix, Steven < <>>
>  the event entity

View raw message