oooh yeah. with sparse impl should be able to handle largeish datasets. the whole n x n matrix would get too large with regular impl.
fastai
