perf: rewrite to native python/Mojo for algorithmic optimizations

[MartinBernstorff]

N: Number of prediction times M: Number of values

• Grouping values and prediction times by patient
  • Probably use Polars for this, seems like something we cannot optimize
• Windowing, use binary-search and sliding windows
  • If we sort prediction times, we know that the start of the window is always increasing
    • O(n log n) for sorting both
    • O(log m) for finding thresholds
    • O(c · m) for copying to each prediction time (where c < 1)
      • Can probably use pointers here
    • O(n) for windowing every prediction time
    • In BigO, no difference, but strong benefits on the constants
• Join aggregated values on prediction times using a dictionary
  • O(1) instead of O(n)

[github-actions[bot]]

This issue is stale because it has been open for 14 days with no activity. Feel free to either 1) remove the stale label or 2) comment. If nothing happens, this will be closed in 7 days.

[github-actions[bot]]

This issue was closed automatically. Feel free to re-open it if it's important.