cosine.go

package metron

import "math"

// CosineSim calculate the similarity of two non-zero vectors using dot product (multiplying vector elements and summing) and deriving the cosine angle between the two vectors.
// TODO not fully tested yet!
//
// Dot Product:
//	  \vec{a} = (a_1, a_2, a_3, \ldots), \vec{b} = (b_1, b_2, b_3, \ldots); where a_n and a_b are elements of the vector.
// Cosine Similarity:
//	  cos(theta) = (A dotProduct B) / (||A|| ||B||)
// nice post: http://blog.christianperone.com/2013/09/machine-learning-cosine-similarity-for-vector-space-models-part-iii/
func CosineSim(a []float64, b []float64) float64 {
	// TODO not fully tested yet!
	var sum, s1, s2 float64
	var count int
	lenA := len(a)
	lenB := len(b)

	if lenA != lenB {
		panic("vectors must be equal length!")
	}

	if lenA > lenB {
		count = lenA
	} else {
		count = lenB
	}
	for k := 0; k < count; k++ {
		if k >= lenA {
			s2 += math.Pow(b[k], 2)
		}
		if k >= lenB {
			s1 += math.Pow(a[k], 2)
		}
		sum += a[k] * b[k]
		s1 += math.Pow(a[k], 2)
		s2 += math.Pow(b[k], 2)
	}
	if s1 == 0 {
		panic("First vector passed is all zeros. Non-zero vectors required.")
	}
	if s2 == 0 {
		panic("Second vector passed is all zeros. Non-zero vectors required.")
	}
	return sum / (math.Sqrt(s1) * math.Sqrt(s2))
}

// CosineDist is cosine distance; simply 1 - CosineSim
func CosineDist(a []float64, b []float64) float64 {
	return 1.0 - CosineSim(a, b)
}