epurdom
use fastcluster for hierarchical clustering
http://danifold.net/fastcluster.html
Should probably be as simple as calling the hclust from this package rather than stats.
Just implemented this in the feature/fastcluster branch.
It was indeed as easy as changing a few lines to use fastcluster::hclust.
Assuming it passes all checks, can I merge into develop?
So apparently fastcluster::hclust() and stats::hclust() do not return the same exact value, because when running the vignette with the same parameters, with stats we get two clusters and with fast cluster only one. This causes an error in getBestFeatures which makes the build fail.
I agree we should wait. Perhaps they just have different defaults that we need to align…
On Apr 13, 2018, at 10:10 PM, Davide Risso [email protected] wrote:
Perhaps we should wait on this since hclust() is not the main bottleneck.
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/epurdom/clusterExperiment/issues/237#issuecomment-381248143, or mute the thread https://github.com/notifications/unsubscribe-auth/AHXGVfWqWCoUvv3l24PeLUrZK1L1OXEYks5toQYagaJpZM4SsIDW.