This is a part of team Villanova's final solution. (22th place) Created by Jiwei Liu and Xueer Chen
It generates solution with 0.90625 public LB
data is first cleaned by removing repeated points in the trip
usage:
python genfea.py # generate features for all drivers sh run.sh # run the model 16 times in parallel python combine.py # average the predictions
It takes about 4 hours in total.