Please note that this is a follow up issue to #4701 and #4197 not a duplicate.

Calling sklearn.metrics.silhouette_samples with a large data set will cause a memory MemoryError in numpy.dot. This is not an issue with numpy.dot but with the implementation of sklearn.metrics.silhouette_samples. For a data set of length N it computes the full NxN pairwise distance matrix which can clearly take a lot of memory. Mathematically all that is needed in memory is a list of the same length as the largest cluster and the rest can be done by appropriate looping.

This issue is to propose a new feature where a new parameter is added to the sklearn.metrics.silhouette_samples method. This would be a boolean to indicate if one wanted to use the faster but more memory consumptive existing method or the alternative method which is better on memory but likely slower.

A working version of the the new method was posted in an answer on StackOverflow