3

We are using Prediction.IO for our recommendation Engine.

After it is deployed (pio deploy), we can call the web service, it has created after deploying, to get the recommendation results one user at a time when needed. We can also send many requests in a single call for many users.

Because we have billions of events and 100s of millions of user entities, using the web service from the http port is not a fast way to export the data and analyse it, say, in Hive.

The problem is, we could not find a way to export the recommendations for all users (let us say 10 recommendations for each user).

There is a command (pio export) to export all the event data, but this does not contain the score values or the recommendation results.

Any thoughts on finding a way to bulk export the results?

Sunil Garg
  • 14,608
  • 25
  • 132
  • 189
myuce
  • 1,321
  • 1
  • 19
  • 29

1 Answers1

1

After some research found this on PredictionIO web site;

This how-to tutorial would explain how you can also use $pio eval to persist predicted result for a batch of queries. Please read the Evaluation to understand the usage of DataSoure's readEval() and the Evaluation component first.

Also;

This tutorial is based on some experimental and developer features, which may be changed in future release.

myuce
  • 1,321
  • 1
  • 19
  • 29