Azure Machine learning designer training and automate batch inference using Azure Synapse and Azure data bricks
2 min readMay 21, 2022
Use end to end batch inference using synapse, azure data bricks and AML batch inference pipeline
Prerequisites
- Azure account
- Azure Machine learning account
- Azure storage account
- Azure databricks account
- Azure synapse workspace account
Architecture
- Using AML Designer to create a batch inference pipeline
- Automate Batch inferencing
Designer Training
- Create a experiment in designer
- Choose computer cluster
- Use open source dataset
- Click Sumbit and train the model
- Select Create batch inference pipeline
- Create a data store to ADLS gen2 with new dataset with empty file.
- Then add export data
- Save the output as parquet and give a filename
- after submit and wait for the run to complete
- then click publish
- Wait for the batch inference endpoint to publish
End to End automated batch inference
- Now go to azure synapse analytics
- Now create a pipeline
- Drag Azure databricks and connect to ADB workspace
- Select the notebook — this creates input batch dataset and stores in batchinput container as parquet file
- Then Drag Azure ML and Select the publish pipeline
- Then drag another Azure databricks and select the notebook to consume batch output and store back in delta table
Output
- Finalize the batch inference pipeline run
Original article — Samples2022/designerdeploy.md at main · balakreshnan/Samples2022 (github.com)