func (p *ActionPredictor) Train(iterations int, dataset DataSet, tokenizer *Tokenizer) {
set := &mlearning.SimpleIterator{}
for _, data := range dataset {
tok := tokenizer.Tokenize(data.Sentence)
set.FeatureSlice = append(set.FeatureSlice, &Meaning{
Tokens: tok,
Vars: data.Vars,
})
set.ClassSlice = append(set.ClassSlice, mlearning.Class(data.Action))
}
for it := 0; it < iterations; it++ {
set.Reset(true)
c, n := p.Perceptron.Train(set)
fmt.Printf("ActionPredictor iter %d: %d/%d=%.3f\n", it, c, n, float64(c)/float64(n)*100)
}
}
func (p *VarPredictor) dataToIterator(dataset DataSet) *mlearning.SimpleIterator {
vs := &mlearning.SimpleIterator{}
for _, data := range dataset {
sen := data.CleanedSentence("[name]")
tok := p.Tokenizer.Tokenize(sen)
for i, t := range tok {
if !strings.HasPrefix(t.Value, "[") {
continue
}
name := strings.Trim(t.Value, "[]")
vs.FeatureSlice = append(vs.FeatureSlice, &varFeature{
Sentence: tok,
Action: data.Action,
Pos: i,
})
vs.ClassSlice = append(vs.ClassSlice, mlearning.Class(name))
}
}
return vs
}
func (p *ActionPredictor) ReinforceMeaning(m *Meaning, action string) {
feats := m.Features()
guess, _ := p.Perceptron.Predict(feats)
p.Perceptron.Update(mlearning.Class(action), guess, feats)
}