search

Paulj1989 / expected-kills

Predicting binary outcomes (kill or error) in volleyball to produce an Expected Kills (xK) model.
MIT License
1 stars 0 forks source link

Incorporate experiment tracking #4

Open Paulj1989 opened 2 years ago

Paulj1989 commented 2 years ago

Is your feature request related to a problem? Please describe. The model development process has been, at times, slow, and this has been made worse by having a less defined monitoring/tracking process.

Describe the solution you'd like Using MLOps tools like DVC, Kubeflow, or MLflow could be used for tracking experiments, as well as offering other potential benefits like version controlling data and, later down the line, easing deployment.

Paulj1989 commented 2 years ago

I've done some initial work to implement DVC, largely to test the process and see if it is a good fit. For the moment it seems worth continuing to explore DVC, but there are some issues deciding where Tidymodels should control the process and what belongs under the DVC umbrella.