Posts on the Alpisto engineering blog filed under #gbm.
The technical story of the Flyability forecast on the FlyBeeper map: five million XContest flights, GFS weather, cumulative XC-distance probabilities, confident learning against self-reported logs, isotonic calibration, and an honest evaluation protocol. From a binary classifier and an attention network to three gradient-boosted trees that ship inside the package and refresh a map layer four times a day.