As spotted by our continuous benchmark suite, there is a more than 2x slowdown in pairwise_distances on sparse input data (for the Euclidean metric).

• https://scikit-learn.org/scikit-learn-benchmarks/#metrics.PairwiseDistancesBenchmark.time_pairwise_distances?p-representation='sparse'&p-metric='euclidean'&p-n_jobs=4&commits=b4afbeee-fabe1606

This happened between b4afbee and fabe160.

Could it be ef5c087?

It's not sure because there are many other commits in git log b4afbeee..fabe1606 and I did not review them all.