Spaces:

---

calibrationcomp

/

calibration_benchmark

Like 0

Running

Added Goal-oriented Bayesian Optimal Experimental Design for the L63, and L96 const + vec runs. Note that the batch optimization is VERY finicky, and very slow, so lots of guards and adjustments are made to make the updates feasible in a reasonable period of time (~24hours with multithreaded parallelization per exp.) and this was not enough for the higher dim experiments. Did not seem worth attempting to further optimize at this time.

Also changed in the leaderboard

• the method taxonomy display, (table)
• the selector mechanism: dropdown -> pills,
• now displays the "top 5 algorithms" by default (as bad algorithms render the plots difficult to read due to axis scales, and frustrating to remove them all)
• more method colors