This demo is based on the original anomaly detection accomplishing the Near Real-Time Anomaly Detection with Delta Live Tables and Databricks Machine Learning blog post.
It uses the kaggle CC dataset to show it can be applied for anomaly detection. The specific components used are
-> The DBR ML runtime (tested with DBR 11.3)
-> DBR managed MLFlow
-> Databricks ML Serving
-> Optimized Pyspark Pandas UDFs for batch scoring.
This repository is a modified version of the original repository and consists of the following files with code:
iforest_train_register.py- quickly trains an isolation forest model and registers it to the model registry. Tweak it as you see fit for your use.json_record_generator.py- shows a way to generate dummy json files should you choose to use the demo data from kaggle.dlt_task_0_iforest_dlt_udf.py- contains code that wrap the model into a pandas udf for scoring inside DLTdlt_task_1_anomaly_detector.sql- shows a simplified end to end DLT workflow that reads the json files created in step 2 and scores that with the trained anomaly detection model registered as a udf on step 3.
To read more about DLT and help getting started with DLT - visit the following link
You can use code in the terraform folder to setup everything automatically. Follow instructions there.