- EDA for ETL
- create synth data generation script
- create pipeline to generate and land new data in cloud storage
- load initial data to cloud storage
- create ETL pipeline for incoming training data
- create ETL pipeline for incoming test/predict data
-
EDA for ML
-
create pipline to load and preprocess data
- feature engineering
- data cleaning
- create funcs for loading pipelines
- create funcs for creating and saving pipelines with different steps
-
create MLOps structure
- set up experiment tracking
- set up metrics logging within modeltests
- set up feature store (discarded)
- set up model registry
-
explore and tune models
- manual exploration
- automl
-
create pipeline for refitting and saving preprocessing Pipeline objects
-
register good models
-
create pipeline for training current best model
-
create prediction pipeline
-
create eval pipeline
-
create pipeline for model re-evaluation/selection
-
create system to monitor performance and trigger re-training and possibly re-selection of new models
- set up precommit
- set up MLOps components in databricks
- put jobs into
.databricks/resources/ - create github action CI/CD that deploys to dev/prod respectively