Feel free to play around with this and the previous version in pandas and see if I have missed something.
Unfortunately there were some intersections between the train and inference split, but not a lot, < 5% if I'm not mistaken.
In these cases, I removed them from the inference split, so testing should be safe now.
Feel free to play around with this and the previous version in pandas and see if I have missed something. Unfortunately there were some intersections between the train and inference split, but not a lot, < 5% if I'm not mistaken. In these cases, I removed them from the inference split, so testing should be safe now.