Calibration method

- Adapt [Dan’s implementation](https://github.com/cmu-delphi/forecasting-team-notebooks/tree/main/quantile-tracker) into an exploration + production script and tune parameters
- Learning rate - .1 in paper, probably somewhere between 0.05 and 0.5
- Window in which to build the bounds - log, 3 or 4 values, 10, 25, 50, all
- Grid across these. Maybe at some point do these independently for each quantile later.
- There’s a for loop that probably just needs to be epix_slide
- Just get best forecaster from David’s experiments and calibrate that
- For now, can just use last year’s forecaster and forecasts to get intuition for tuning parameters
- [Lengthy Slack thread](https://delphi-org.slack.com/archives/C01D9R5QDT7/p1729223455013329)