search

nolanlab / spade

SPADE: Spanning Tree Progression of Density Normalized Events
Other
46 stars 23 forks source link

The change of the clustering distance #130

Open SamGG opened 7 years ago

SamGG commented 7 years ago

Hi, When SPADE switches from the original clustering (cluster.cc) to the Rcppcluster one, it seems that the distance changed from Manhattan to Euclidean also. What is the rational? Best.

monchin commented 6 years ago

@SamGG It seems that when in driver .R to call FCSToTree function, it would call SPADE.cluster function and then call Rclusterpp.hcluster function. In Rclusterpp.hcluster, the default parameter is Rclusterpp.hclust <- function(x, method="ward", members=NULL, distance="euclidean", p=2) which means the distance is Euclidean. But I'm wondering why does SPADE compute the Manhattan distance when it compute density and upsampling, while use Euclidean distance and 'Ward' method when it h-clusters. I don't know the reason to use Manhattan.