// Xmeans runs k-means for k lower bound to k upper bound on a data set.
// Once the k centroids have converged each cluster is bisected and the BIC
// of the orginal cluster (parent = a model with one centroid) to the
// the bisected model which consists of two centroids and whichever is greater
// is committed to the set of clusters for this larger model k.
//
func Xmeans(datapoints, centroids *matrix.DenseMatrix, kmax int, cc, bisectcc CentroidChooser, measurer VectorMeasurer) ([]Model, map[string]error) {
logname := "/var/tmp/xmeans.log"
fp, err := os.OpenFile(logname, os.O_RDWR|os.O_APPEND, 0666)
if err != nil {
if os.IsNotExist(err) {
fp, err = os.Create(logname)
if err != nil {
fmt.Printf("Xmeans: cannot open %s for logging.\n", logname)
}
}
}
log.SetOutput(io.Writer(fp))
k, _ := centroids.GetSize()
log.Printf("Start k=%d kmax=%d\n", k, kmax)
R, M := datapoints.GetSize()
errs := make(map[string]error)
runtime.GOMAXPROCS(numworkers)
models := make([]Model, 0)
for k <= kmax {
log.Printf("kmeans started k=%d\n", k)
model, err := kmeans(datapoints, centroids, measurer)
if err != nil {
errs[strconv.Itoa(k)] = err
}
// Bisect the returned clusters
log.Println("bisect started")
bimodel := bisect(model.Clusters, R, M, bisectcc, measurer)
numCentroids := len(bimodel.Clusters)
log.Printf("bisect returned %d clusters\n", numCentroids)
models = append(models, model)
var cent *matrix.DenseMatrix
if numCentroids <= kmax {
for rowexists := true; rowexists == true; {
cent = cc.ChooseCentroids(datapoints, 1)
rowexists = centroids.RowExists(cent)
}
centroids, err = centroids.AppendRow(cent)
if err != nil {
log.Printf("AppendRow: %v\n", err)
errs["ApppendRow"] = err
break
}
k++
} else {
k = numCentroids
}
}
log.Println("Finished")
return models, errs
}
// Xmeans runs k-means for k lower bound to k upper bound on a data set.
// Once the k centroids have converged each cluster is bisected and the BIC
// of the orginal cluster (parent = a model with one centroid) to the
// the bisected model which consists of two centroids and whichever is greater
// is committed to the set of clusters for this larger model k.
//
func Xmeans(datapoints, centroids *matrix.DenseMatrix, k, kmax int, cc, bisectcc CentroidChooser, measurer VectorMeasurer) ([]Model, map[string]error) {
var err error
// Uncomment logging code as well as the import statement above if you want simple logging to the elapsed
// time between major events.
/* logname := "/var/tmp/xmeans.log"
fp, err := os.OpenFile(logname, os.O_RDWR|os.O_APPEND, 0666)
if err != nil {
if os.IsNotExist(err) {
fp, err = os.Create(logname)
if err != nil {
fmt.Printf("Xmeans: cannot open %s for logging.\n", logname)
}
}
}
log.SetOutput(io.Writer(fp))
*/
if k > kmax {
m := make([]Model, 0)
e := map[string]error{
"k": errors.New(fmt.Sprintf("k must be <= kmax. Received k=%d and kmax=%d.", k, kmax)),
}
return m, e
}
// log.Printf("Start k=%d kmax=%d\n", k, kmax)
R, M := datapoints.GetSize()
errs := make(map[string]error)
runtime.GOMAXPROCS(numworkers)
models := make([]Model, 0)
for k <= kmax {
// log.Printf("kmeans started k=%d\n", k)
model := kmeans(datapoints, centroids, measurer)
// Bisect the returned clusters
// log.Println("bisect started")
bimodel := bisect(model.Clusters, R, M, bisectcc, measurer)
numCentroids := len(bimodel.Clusters)
// log.Printf("bisect returned %d clusters\n", numCentroids)
models = append(models, model)
var cent *matrix.DenseMatrix
if numCentroids <= kmax {
for rowexists := true; rowexists == true; {
cent = cc.ChooseCentroids(datapoints, 1)
rowexists = centroids.RowExists(cent)
}
centroids, err = centroids.AppendRow(cent)
if err != nil {
errs["ApppendRow"] = err
break
}
k++
} else {
k = numCentroids
}
}
// log.Println("Finished")
return models, errs
}