Predictionio Buildpack Versions Save

Deploy predictive query engines built with PredictionIO, an open-source machine learning framework.

v10.0.0

6 years ago

This buildpack now supports the official 0.12.0-incubating release of PredictionIO. For the first time since this project began, it no longer requires a special patched version to work on Heroku. The buildpack now downloads the binary distribution directly from Apache Software Foundation.

To use this version, see updating source configs.

The major changes in PredictionIO include:

v9.0.1

6 years ago

Documentation revision alerting loss of support for PredictionIO 0.10.0-incubating.

v9.0.0

6 years ago

Support for Buildpack Composition ?

? Potentially breaking change. See details in the original pull request.

This version removes an old filesystem path workaround that is no longer required for Heroku compatibility.

Now, this buildpack may be composed with other buildpacks, like Python or Node. This is most useful for creating data flow scripts.

v8.3.0

6 years ago

? This is the last buildpack version to support PredictionIO 0.10.0-incubating.

Batch Predictions ?

We pull requested a new pio batchpredict command for PredictionIO and added it in the 0.11.0-SNAPSHOT distribution that's included with this buildpack. See the pull request for more details and how to use pre-release features.

Fixes

v8.2.0

6 years ago

Local Development

Heroku deployment

v8.1.0

6 years ago

v8.0.0

6 years ago

New things for the PredictionIO on Heroku developer!

Local Development Workflow ???

? See the Local Development docs for complete how-to.

After fumbling around with a dozen local PredictionIO projects, the pain of managing differing pio-env.sh and storage configs along with various PredictionIO versions just became too much ? This buildpack already performs a complete setup of PredictionIO on every Heroku deploy, so we decided to extend that automated setup to our local computers.

Powered by two commands, PredictionIO may now be installed into each individual engine directory, very much like git push heroku master on Heroku:

$PIO_BUILDPACK_DIR/bin/local/setup

That command installs everything needed to run an engine into PredictionIO-dist/ directory, and it installs the script:

bin/pio

Use this for PredictionIO's CLI, such as bin/pio status & bin/pio train. It reads the local environment (config vars) from the local .env file everytime it's invoked and then calls through to the real pio command.

Setup for local development is an order of magnitude simpler for folks on a supported platform, and this automated local setup ensures the local dev system matches Heroku deployment, avoiding differences that cause bugs and insanity.

Self-contained Buildpack ?

No longer do PredictionIO engines & eventservers require a stack of multiple buildpacks. Just a single setting:

heroku buildpacks:set https://github.com/heroku/predictionio-buildpack

⚠️ For pre-existing apps, this means heroku/jvm-common or heroku/scala should be removed. Replaced with only https://github.com/heroku/predictionio-buildpack.

# View current buildpacks
heroku buildpacks

# Clear all buildpacks
# (unless other additional buildpacks are present/needed)
heroku buildpacks:clear

# Set PredictionIO buildpack
heroku buildpacks:set https://github.com/heroku/predictionio-buildpack

Improve Spark memory config

It's now possible to train a model with a small dataset on a free dyno w/ 512MB RAM, like this classification engine (7aef127b53cc0d7c32f7153aecf9419392cf9f98)

Improve Postgres config

Enabling indexes and specifying concurrency to optimize storage performance. (8492053022d051e4178b5979bc530adf117b570f)

Fixes

  • PIO_TRAIN_ON_RELEASE=false now makes the release process a no-op (eacfc2aa27a9b217d4293920889eb3ffc484a354)
  • Mistakes from transition to PredictionIO 0.11.0-incubating (c5b138ff9fa245f63675197edb66a11dd4fa0e7f, 87915ba8b4ccfcf5850bc82015e35b866babf42f)

Thanks ?

Please use this repo's issues to provide feedback or search for and report bugs.

v7.1.0

7 years ago

? Experimental support for authenticated Elasticsearch 5.1.1 using Bonsai, a Heroku Add-on.

? Enables deployment of the Universal Recommender using this UR fork which includes its own revisions to work with Elasticsearch 5. Also requires this PredictionIO fork which implements optional basic HTTP auth to Elasticsearch 5.

  • set PIO_ELASTICSEARCH_URL to activate Elasticsearch 5 as the metadata store

  • requires Elasticsearch 5.1.1

    heroku addons:create bonsai:shared-10 --version 5.1 --as PIO_ELASTICSEARCH
    
    • versions are not interchangeable
    • the shared-10 plan is required to provide adequate concurrency for the Universal Recommender's Spark-Elasticsearch integration
  • build.sbt & template.json must depend on PredictionIO 0.11.0-SNAPSHOT, a distribution provided in this buildpack's local Maven repo/.

v7.0.0

7 years ago

Support for deploying PredictionIO 0.11.0-incubating:

  • exclusive support for Scala 2.11.8, Spark 2.1.0, & Hadoop 2.7.3 with version 0.11.0-incubating
  • build is now compatible with Private Spaces & Private Databases
  • S3 HDFS is now compatible Amazon S3 buckets that require signatureV4 (regions newer than mid-2014)
  • new, optional data import & synchronization hooks that automatically run during release, before training (documentation forthcoming)

v6.0.0

7 years ago

Improve behavior (breaking change) of custom data loader script to feed the automated pio import command.

See Import data docs.