Thanks- if we get it working in a disributed configuration I’ll report back how.
On Tue, Apr 2, 2019 at 12:59 PM Umesh kaushik <umesh.kaus...@popxo.com> wrote: > Hi Jonathan, > > I am using predictionion in production. > Their documentation for distributed cluster is screwed up. Hbase version > they have mentioned is not compatible with the Hadoop version they have > mentioned. I tried make it distributed but could not (did not give much try > either), so we are using it on single machine and developing our own > recommendation engine along side it. Event server lies on one machine, on > that it will manage multiple apps (if you deploy more than one app). > Apart from that it's decent in handling requests. Since I could not make > it distributed so using r5d.12xlarge for training and r5d.2x large for > normal processing. > > Cheers, > Umesh Kaushik > (9620023458) > > sent from a mobile device, please ignore the typographical errors. > > On Tue, 2 Apr, 2019, 10:42 PM Jonathan Barlow, <j...@barlownet.com> wrote: > >> Hi All, >> >> I'm new to the list. I apologize if this question has been asked before. >> >> Eventually, our goal is to evaluate PredictionIO to serve as a platform >> inside of a research organization. Our data scientists will develop engines >> and we will deploy them to PredictionIO. After training, we would like to >> hit REST endpoints in PredictionIO to add new training data and we would >> like to be able to hit REST endpoints to get predictions (example, sending >> some text to the prediction server for classification results). >> >> I would like to evaluate PredictionIO and would like to get your advice >> on an actual deployment architecture. >> >> The architecture section on the PredictionIO website is fairly abstract. >> I can't determine if, for example, the EventServer represents a cluster, a >> single machine, etc. >> >> It looks like, if I want the minimum configuration with three nodes >> supporting hbase, it would be something like this: >> >> Server1: >> Hadoop >> Hbase >> Spark >> ElasticSearch >> PredictionIO (Events?) >> >> Server 2: >> Hadoop >> Hbase >> ElasticSearch >> Spark >> PredictionIO (For Predictions?) >> >> Server 3: >> Hadoop >> Hbase >> ElasticSearch >> Spark >> >> Or should I put PredictionIO on a fourth server outside of the hadoop >> cluster? Or should I have two Prediction IO servers outside of the hadoop >> cluster, one for Events and one for Predictions? >> >> I also wonder about scaling - how would we scale the prediction server if >> we started hitting the prediction endpoints a lot in from a production >> server? >> >> Anyway, I appreciate any help! >> >> Thanks, >> Jonathan Barlow >> >> >> >> >> >> >> >> >> >>