func main() {
sumR := 0.0
sum3 := 0.0
sum5 := 0.0
sum10 := 0.0
for _, importer := range load.Ambiant {
corpus := importer.Import()
dpmom := cluster.NewDPMoM(corpus, 1, 20, .2, .2)
graphical.RunCCM(dpmom, cluster.NewConvergenceCheck(dpmom.Z), 0)
gold := importer.Label(corpus)
pred := eval.NewClusteringState(dpmom.Z)
cont := eval.NewContingency(gold, pred)
rerank := eval.Rerank(dpmom.Z)
sumR += cont.Rand()
sum3 += eval.SRecallK(3, rerank, gold)
sum5 += eval.SRecallK(5, rerank, gold)
sum10 += eval.SRecallK(10, rerank, gold)
}
n := float64(len(load.Ambiant))
fmt.Println("Rand :", sumR/n)
fmt.Println("SR@3 :", sum3/n)
fmt.Println("SR@5 :", sum5/n)
fmt.Println("SR@10:", sum10/n)
}
func (r Runner) EvalQuery(importer load.Importer) Stats {
corpus := importer.Import()
gold := importer.Label(corpus)
z := r(corpus)
pred := eval.NewClusteringState(z)
cont := eval.NewContingency(gold, pred)
rerank := eval.Rerank(z)
stats := Stats{}
stats.Rand = cont.Rand()
for i, k := range Ks {
stats.Recall[i] = eval.SRecallK(k, rerank, gold)
}
return stats
}
func main() {
rand.Seed(time.Now().UnixNano())
corpus := load.Newsgroups.Import()
gold := load.Newsgroups.Label(corpus)
mom := cluster.NewMoM(corpus, 20, .1, .01)
checker := cluster.NewConvergenceCheck(mom.Z)
converged := false
var dur time.Duration
var iter int
fmt.Println("iter\ttime\tchanges\tf-measure\tARI\tVI")
for !converged {
start := time.Now()
mom.CCM()
end := time.Now()
dur += end.Sub(start)
iter += 1
pred := eval.NewClusteringState(mom.Z)
cont := eval.NewContingency(gold, pred)
changes := checker.Check()
line := []string{
fmt.Sprintf("%d", iter),
fmt.Sprintf("%.3f", dur.Seconds()),
fmt.Sprintf("%d", changes),
fmt.Sprintf("%.3f", cont.FMeasure()),
fmt.Sprintf("%.3f", cont.ARI()),
fmt.Sprintf("%.3f", cont.VI())}
fmt.Println(strings.Join(line, "\t"))
converged = changes == 0
}
fmt.Println()
for z := 0; z < mom.T; z++ {
fmt.Printf("%d: %s\n", z, mom.TopicSummary(z, 10))
}
}
