Operations
Running the project and shipping its results: execution environments, the artifacts a run produces, experiment tracking, Kaggle packaging and documentation publishing.
1. What this section covers
Development pages explain how to change the code. This section explains how to run it and where its artifacts go: what a training run leaves on disk, how to pick a run back up, how predictions reach a submission, and how this documentation is published.
| Question | Answered in |
|---|---|
| Where do checkpoints, logs and resolved configs go? | Deployment — Output layout |
| How do I resume or compare runs? | Deployment — Checkpoints |
| How does an experiment reach the leaderboard? | Deployment — Kaggle packaging |
| How is this site published? | Deployment — Publishing the docs |
Environment-specific paths Several entry points carry absolute paths from the machine the pipeline was developed on. Treat any path in a default argument as a placeholder to override, not as a working default, and pass explicit paths on the command line. Deployment summarises which settings to check first when moving to a new machine.