package/kmeans
1
0
Fork 0
mirror of https://github.com/muesli/kmeans.git synced 2025-09-27 03:56:17 +08:00
Code Issues Packages Projects Releases Wiki Activity

Don't empty one cluster just to fill another

Browse Source
This commit is contained in:
Christian Muehlhaeuser
2018-05-29 03:57:24 +02:00
parent fd3d39c8f6
commit 86f284616f
1 changed files with 13 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
kmeans.go
View File

@@ -90,17 +90,23 @@ func (m Kmeans) Partition(dataset Points, k int) (Clusters, error) {
} }
} }
for ci, c := range clusters { for ci := 0; ci < len(clusters); ci++ {
if len(c.Points) == 0 { if len(clusters[ci].Points) == 0 {
// During the iterations, if any of the cluster centers has no // During the iterations, if any of the cluster centers has no
// data points associated with it, assign a random data point // data points associated with it, assign a random data point
// to it. // to it.
//
// Also see: http://user.ceng.metu.edu.tr/~tcan/ceng465_f1314/Schedule/KMeansEmpty.html // Also see: http://user.ceng.metu.edu.tr/~tcan/ceng465_f1314/Schedule/KMeansEmpty.html
ri := rand.Intn(len(dataset)) var ri int
c.Points = append(c.Points, dataset[ri]) for {
// find a cluster with at least two data points, otherwise
// FIXME: remove Point from previously assigned cluster? // we're just emptying one cluster to fill another
ri = rand.Intn(len(dataset))
if len(clusters[points[ri]].Points) > 1 {
break
}
}
clusters[ci].Points = append(clusters[ci].Points, dataset[ri])
clusters[points[ri]].removePoint(dataset[ri])
points[ri] = ci points[ri] = ci
} }
} }