best_centers update problem

[hjcen]

scikit-learn/sklearn/cluster/k_means_.py

Line 559 in 13981bd

best_centers = centers.copy()

This line should be best_centers = centers_old.copy() ?
Because the inertia in code L541 is calculated for old centers.

[jnothman]

Your opinion, @jeremiedbb?

[jeremiedbb]

I think the existing code is correct. In the E step, we compute the labels and inertia based on the previous centers and in the M step we compute the new centers based on the labels.

Also, after the iterations loop, we rerun the E step once more so that the labels match the final centers. So in the end the labels and inertia are computed on the last new centers.

[jeremiedbb]

Outdated by #11950. Closing.