Productionizing H2O Models with Apache

Spark

Jakub Háva,jakub@h2o.aihttps://github.com/jakubhavahttps://www.linkedin.com/in/havaj/

AI UkraineKyiv, October 13-14 2018

#ML4SAIS

Who are we?• Kuba

• Senior Software engineer at H2O.ai - Core Sparkling Water • Master’s at Charles University (CZ) • Implemented high-performance cluster monitoring tool for

JVM based languages (JNI, JVMTI, instrumentation) • Michal

• VP of Engineering at H2O.ai • Creator of Sparkling Water • Ph.D at Charles University (CZ), PostDoc at Purdue

!2

Machine Learning (ML) Lifecycle

!4

ModelTrainingAlgorithm

FeatureEngineering

ModelPipelineBuilding

TrainingPredictions

DataEngineering

Basic ML Lifecycle

#ML4SAIS

!5

ModelTrainingAlgorithm

FeatureEngineering

FeaturizationPipelineModel

ModelPipelineBuilding

TrainingPredictions

DeploymentPredictions

DataEngineering

ModelPipelineDeployment

Basic ML Lifecycle

#ML4SAIS

Example Implementations

!6#ML4SAIS

Data Engineering Feature Engineering Training Algorithm Deployment

Pipeline Model

Spark H2O Spark H2O MOJO

Spark H2O Driverless AI Spark H2O Driverless AI MOJO

Model Building Model Deployment

H2O + Spark = Sparkling

Water

#ML4SAIS

H2O + Spark• H2O

• Machine Learning Library • Distributed Algorithms • For ML experts

• Sparkling Water • Integrates H2O & Spark Ecosystems • Transparent for Spark users • Based on Spark pipelines & H2O

!8

Basic ML Lifecycle: Sparkling Water

!9

ModelTrainingAlgorithm

FeatureEngineering

SparkTransformers H2OMOJOModel

TrainingPredictions

DeploymentPredictions

AutoML

Pipeline

#ML4SAIS

Demo: Spark Pipeline

H2O Driverless AI

#ML4SAIS

H2O Driverless AI• What if I’m not expert ?

• H2O Driverless AI • H2O Driverless AI

• No expert knowledge required • Automatic Feature Engineering & ML

!12

Basic ML Lifecycle: Driverless AI

!13

ModelTrainingAlgorithm

FeatureEngineering

DriverlessAIFeatureTransformations DriverlessAIModel

TrainingPredictions

DeploymentPredictions

PipelineDriverlessAIMOJOas

#ML4SAIS

Demo: Driverless AI as Spark Pipeline

!15

Driverless AI Pipeline

!16#ML4SAIS

Governed ML Lifecycle

Governed ML Lifecycle

!18

ModelTrainingAlgorithm

FeatureEngineering

FeaturizationPipelineModel

ModelPipelineBuilding

TrainingPredictions

DeploymentPredictions

ModelManagement

DataEngineering

ModelPipelineDeployment

ModelMonitoring

AutoDocumentation

#ML4SAIS

#ML4SAIS

Materials

!19

https://bit.ly/2sxowxD

#ML4SAIS

Sparkling Water enables deployment of H2O ML models with Spark Pipelines

!20

Thank you!