Introducing apache prediction io (incubating) (bay area spark meetup at salesforce)

IntroducingApache PredictionIO (incubating)http://predictionio.incubator.apache.org

Donald Szeto

Tech Lead @ SalesforceFirst Git Commit @ PredictionIO

Agenda

Future {

println(roadmap)

}

?

(Source: benbrandt22, https://redd.it/383edw)

Why PredictionIO?

ML problem 1 Data collection Modeling Serving/scoring

Evaluation

ML problem 2 Data collection Modeling Serving/scoring

Evaluation

Is there a common factor?(Not his actual words :))

What is PredictionIO?

A machine learning server for developers and ML engineers

PredictionIO API

Engines EnginesEngines

Quick Demo

Digging Deeper

What is DASE?Data, Algorithm, Serving, Evaluation

What are engine instances?

What are engine variants?

Engine Instances

Engine(Scala/Java code)

Data

Engine Parameters

Algorithm Hyperparameters

Environment

+ Engine Instance=

Digging Deeper

What is DASE?Data, Algorithm, Serving, Evaluation

What are engine instances?

What are engine variants?

Engine Variants

Engine(Scala/Java code)

Data

Engine Parameters

Algorithm Hyperparameters

Environment

+ Engine Instance=

Data

Engine Parameters

Algorithm Hyperparameters

Environment

Engine Instance

Current Development ( <= 0.10.0 )

Migrating to ASF infrastructure

Merging forksSliding window event data source

Installation fixes

Engine templates and SDKs migration

Docker-based Integration Test Infrastructure

Travis CI Worker

Travis CI Worker

Travis CI Worker

Travis CI Worker

Docker-based Integration Test Infrastructure

Running many test engines in different environments, in parallel

Future Roadmap ( > 0.10.0 )

Cross-building with Spark 1.x and Spark 2.x

Better native support of Spark ML Pipeline and DataSet

Multi-engine serving

Admin API w/ CLI Refactoring

Testing infrastructure for community engine templates

Further separation-of-concernsAllow engines to use different parts of PredictionIO independently

WE NEED YOUR HELP!!!Please subscribe to [email protected]

For usage questions please subscribe to [email protected]

donald (at) apache.org

Thank you!