A repo for FEAST debugging scripts for a test task (some feast tutorials are ported to Apache Hive offline store)
Our sandbox has batteries included: we provide local Docker servers of Redis and Apache Hive. To start them, type
bash start_docker.sh
To stop, type
bash stop_docker.sh
feast_sandbox/driver_tutorial/download_data.py: downloads data table with driver data from BigQuery
and stores it to repos/driver_parquet_repo folder as parquet file .
feast_sandbox/driver_tutorial/populate_hive_base.py: uploads previously downloaded data table to Apache Hive
(driver_tuitorial database is created)
feast_sandbox/driver_tutorial/predict_driver.py: main tutorial script
(predicts best driver from 5 given)
A simplified version of normal fraud detection tutorial from Feast
(we infer the AI model locally instead of sending model and features to Google AI platform)
feast_sandbox/fraud_tutorial/download_data.py: downloads data table with fraud data from BigQuery
and stores it to repos/fraud_parquet_repo folder as parquet files.
feast_sandbox/driver_tutorial/populate_hive_base.py: uploads previously downloaded data table to Apache Hive
(fraud_tuitorial database is created)
feast_sandbox/driver_tutorial/detect_fraud.py: main tutorial script
(trains a classification model and predicts if a transaction is fraudulent)
TODO: port backfill_features function!
Not implemented yet. Maybe later : 3