func edist(sent1, sent2 string, m *wordmodel.Model) C.float {
words1 := strings.Split(sent1, " ")
words2 := strings.Split(sent2, " ")
vec1 := [][]float32{}
vec2 := [][]float32{}
for _, w1 := range words1 {
vec := wordmodel.Vector(make([]float32, m.Layer1Size))
if wordId, ok := m.Vocab[w1]; !ok {
fmt.Printf("word not found: %s\n", w1)
} else {
vec.Add(1, m.Vector(wordId))
}
vec.Normalize()
vec1 = append(vec1, vec)
}
for _, w2 := range words2 {
vec := wordmodel.Vector(make([]float32, m.Layer1Size))
if wordId, ok := m.Vocab[w2]; !ok {
fmt.Printf("word not found: %s\n", w2)
} else {
vec.Add(1, m.Vector(wordId))
}
vec.Normalize()
vec2 = append(vec2, vec)
}
f1, f2 := make([]C.feature_t, len(vec1)), make([]C.feature_t, len(vec2))
for i, v := range vec1 {
f1[i].arr = (*C.float)(&v[0])
}
for i, v := range vec2 {
f2[i].arr = (*C.float)(&v[0])
}
wt1, wt2 := make([]C.float, len(vec1[0])), make([]C.float, len(vec2[0]))
for i := 0; i < len(wt1); i++ {
wt1[i], wt2[i] = 1.0, 1.0
}
s1 := C.signature_t{n: (C.int)(len(words1)), Weights: &wt1[0], Features: &f1[0]}
s2 := C.signature_t{n: (C.int)(len(words2)), Weights: &wt2[0], Features: &f2[0]}
var flow C.flow_t
var flowsize C.int
emd_dist := C.emd(&s1, &s2, (*[0]byte)(C.dist), &flow, &flowsize)
return emd_dist
}
func mostSimilar(query string, n int, m *wordmodel.Model) ([]Pair, error) {
vec := wordmodel.Vector(make([]float32, m.Layer1Size))
if wordId, ok := m.Vocab[query]; !ok {
return nil, fmt.Errorf("word not found: %s", query)
} else {
vec.Add(1, m.Vector(wordId))
}
vec.Normalize()
r := make([]Pair, n)
for w, i := range m.Vocab {
sim := vec.Dot(m.Vector(i))
this := Pair{w, sim}
for j := 0; j < n; j++ {
if this.Sim > r[j].Sim {
this, r[j] = r[j], this
}
}
}
return r[1:], nil
}