Open jkomoros opened 9 years ago
A weaker form of this is "tend to pick cells from {group} that have more cells filled", where block > row > col
When testing whether this is a good change, one way to do it is to take, say, every 10th row of relative_difficulties.csv and train a model on that, make the change, and train a model and see if r2 is better or worse.
Also, from rows within blocks that have more filled.
This might be captured by the normal tendency to pick obvious in block, etc.
I notice myself doing this very often when solving a puzzle. This feels similar to chainDissimilarity