Skip to content

Latest commit

 

History

History
53 lines (33 loc) · 2.79 KB

upgrading_from_pio_to_harness.md

File metadata and controls

53 lines (33 loc) · 2.79 KB
Raw

Upgrading from PIO+UR to Harness+UR

If you are primarily using the Universal Recommender with PIO you should find upgrading fairly easy. Outline:

  1. With PIO running export your data from the UR
  2. start Harness
  3. add a UR engine instance and parameters
  4. import your data
  5. train the UR

After step 3 you are ready for input and after 5 you are ready for queries.

Backup Your Data

With PredictionIO running, export all your data. This is done with:

  • pio app list to find your appid
  • pio export --appid <id-from-step-1> --output </path/to/export/location> export all data

This gives you a snapshot of your PIO hosted data.

Install and Start Harness

Follow instructions to Install Harness. This can be done by hand on your host or by using Docker containers. Start Harness, it already contains the UR and other Engines.

Configure the UR

Harness is fully multi-tenant so every reference to an Engine Instance uses an engine-id. This is created by setting it in a config JSON file with other parameters. The format is similar to the PIO engine.json for the UR with some changes:

  • indicators instead of eventNames
  • rules instead of fields
  • the Harness version also has a slightly different structure.

Create a JSON file for the UR substituting your values from the PIO version of engine.json to fit the description in UR Configuration

Start Your UR Instance

Harness can run many instances of Engines like the UR. Once you have installed Harness and the Harness-cli you can add one that will work like your PIO UR:

  • hctl status checks only harness running status, so hand check that MongoDB and Elasticsearch are also running
  • hctl add </path/to/ur/config/json> This will create an Engine Instance for the UR with the engineId named in the JSON file
  • hctl status engines <ur-instance-id> see the params you set for this instance and verify everything is running.
  • hctl import <ur-instance-id> </path/to/export/location> import from data exported by PIO + UR.
  • hctl train <ur-instance-id> when completed, this will allow you to query the UR using REST or the new Harness SKD (Java, Scala, Python).

Input and Queries

At this point Harness and the UR are ready for live input and queries. To make them you need to use the REST endpoints. Below <harness-host> is your IP address for the instance of Harness you started.

  • POST http://<harness-host>:9090/engines/<ur-engine-instance>/events where the request is a UR input JSON packet that is identical to PredictionIO, only the URI is different.
  • POST http://<harness-host>:9090/engines/<ur-engine-instance>/queries where the request is a UR query JSON packet identical to PIO, only the URI is different.

Alternatively you can use one of the Harness SDKs to do queries and input.