ML iOS Deployments

ML iOS Deployments

Pretext: Aiming for a single ML model to forecast across all USGS stations to avoid the costs associated with training and hosting through third-party services.

Deployment Steps:

1. Model Training:

   • Train a single model
   • Our training CI pipeline outputs a .h5 model file.

2. Model Conversion and Versioning:

   • Use GitHub Actions to automatically convert the .h5 Keras model to CoreML format.
   • Increment the app's version number based on the latest release.

3. Release Packaging:

   • GitHub Actions packages the CoreML model and creates a new GitHub release with an incremented version number.

4. Release Distribution:

   • The iOS app checks for the "latest" release on GitHub.
   • Once found, it automatically fetches and integrates the new CoreML model.