Forecasting With H20

In order to ensure that we are able to give our customers a reflection of both past and future state - it is imperative that we add forecasting capabilities to our product. We will use H20 as the AI/ML platform for forecasting capabilities and it would be integrated to Insights for training and model deployment.

Pre-Requisite:

• Start the H2O jar locally using below command

• Download H20 jar with version h2o-3.30.1.2, Link for download h2o-3.30.1.2.zip

• java -jar h2o.jar 

• By default H2O starts in port: 54321.

• Mention below configuration inside server-config.json.

• "mlConfiguration": {"h2oEndpoint": "http://localhost:54321"}.

• DB Table Entry: INSIGHTS_WORKFLOW_TYPE  :

  • Manual entry in PostgreSQL, mention type as AUTOML . 

  • API Used:- /PlatformService/insights/workflow/saveWorkflowTask  
    Workflow task Json
    {"description":"H2O_AutoML_Execute","mqChannel":"WORKFLOW.TASK.AUTOML.EXCECUTION","componentName":"com.cognizant.devops.automl.task.core.AutoMLSubscriber","dependency":-1,"workflowType":"AUTOML"}

Configure Usecase :

• Navigate to Forecasting: Here you will see already created usecases.

Open

• Click on add button to upload data.

  • Input unique usecase name and upload test csv data.

  • Once the file gets uploaded then below details needs to be filled.

Open

•  Split Ratio and Target Column

• Split Ratio: Select split percentage as Training and Test.

• Maximum Models to run: Specify the number which indicates maximum algorithms to run against training data.

• Response Column: select response column as prediction column.

• Prediction Type: Select prediction type as either Classification or Regression.

• Column Mapping Detail

  • Fill the details for each column type and mention columns used for word2vector analysis.

• Click on the save button to save the usecase.

• The usecase is save as ONETIME workflow and is picked up by Immediate Workflow Scheduler.

Executing Usecase:

• To execute the usecase run the workflow engine.

• Once workflow ran successfully check the status of the workflow as LEADERBOAD_READY as highlighted below.

• Select this usecase and click on the Leaderboard button on the Top.

• Select the best MOJO in the list and Save the MOJO.

• Once MOJO gets saved the Status is changed to MOJO_DEPLOYED as highlighted below and the usecase would be ready for further use in Report Section.

   

KPI Creation:
Refer below sample KPI creation window.

• Category must be selected as PREDICTION.

• Select the usecase from the list. The list of those usecase available whose status is MOJO_DEPLOYED in Usecase Management Screen.

• Save the KPI.

Report Configuration: 

• Report configuration steps are same as Report Management so refer Report Configuration Documentation.

• While creating the report make sure selecting KPI's created at KPI creation step.