Neural forecasting and Python/Go inference are merged — contributors welcome

[vrajpatell]

OpenTrace ML’s latest updates are merged into main.

What’s new:

• Experimental CPU neural forecasting in Python.
• Persistence and seasonal baselines with chronological backtesting.
• Portable linear-model JSON: train in Python and run inference in Go.
• Per-class detection metrics with standard Python mapping support.

The project remains pre-alpha. The initial public-data benchmark covered one UCI traffic window; neural MAE improved over the seasonal baseline, but seasonal MAPE was better and all neural fits reached the iteration limit. Neural export to Go is still open work.

Contributors can start with:

• Evaluate additional public-data windows
• Investigate online-learning behavior
• Design portable neural inference
• Build a tiny offline OSM fixture

Repository: https://github.com/vrajpatell/opentrace-ml

Which experiment or integration would make this library useful in your project?