I broke some data into a training and a hold out set. Then I clustered the training set with the k-means method. Now I want to calculate error using the holdout set. Do I just take the square the distance of each point in the holdout set from the closest centroid, and then sum the distances?
$\begingroup$
$\endgroup$
2
-
$\begingroup$ that is the formula for within-cluster sum of squares (WSS). what exactly are you trying to do? This is not a regression model -- it is not like a low WSS means valid clusters -- if k=n then WSS = 0. There are other techniques you should be using if you are trying to assess the validity of a clustering. $\endgroup$– C8H10N4O2Commented Oct 30, 2015 at 15:48
-
$\begingroup$ I'm trying to use the holdout set productively. Not sure how. $\endgroup$– Damian SiniakowiczCommented Oct 30, 2015 at 15:51
Add a comment
|