func (i *ITM) SampleTopN(n int) {
for _, d := range rand.Perm(i.M) {
for w := 0; w < i.N[d]; w++ {
i.unsetZ(d, w)
allCounts := i.conditional(d, w)
topCounts := make([]float64, i.T)
for _, z := range util.TopN(util.FloatCounts(allCounts), n) {
topCounts[z] = allCounts[z]
}
z := util.SampleCounts(topCounts)
i.setZ(d, w, z)
}
}
}
// TopNPerDoc filters words which are not in the top n in at least one document.
func TopNPerDoc(c *Corpus, n int) *Corpus {
keep := make(map[string]struct{})
for d := 0; d < c.M; d++ {
counts := make([]int, c.V)
for n := 0; n < c.N[d]; n++ {
counts[c.W[d][n]]++
}
for _, v := range util.TopN(util.IntCounts(counts), n) {
keep[c.Vocab.Tokens[v]] = struct{}{}
}
}
remove := make(map[string]bool)
for token := range keep {
if _, ok := keep[token]; !ok {
remove[token] = true
}
}
return FilterStopwords(c, remove)
}
func (b *base) TopicSummaryIds(z, n int) []int {
return util.TopN(util.IntCounts(b.TopicWords[z]), n)
}
// TopicSummaryIds gets the top n tokens ids for the zth topic.
func (i *ITM) TopicSummaryIds(z, n int) []int {
return util.TopN(util.IntCounts(i.TopicConstraints[z]), n)
}
// TopicSummaryIds gets the top n word token ids for the zth topic.
func (l *LDA) TopicSummaryIds(z, n int) []int {
return util.TopN(util.IntCounts(l.TopicWords[z]), n)
}