Spaces:
Running
Running
File size: 4,402 Bytes
b110593 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 |
// _ _
// __ _____ __ ___ ___ __ _| |_ ___
// \ \ /\ / / _ \/ _` \ \ / / |/ _` | __/ _ \
// \ V V / __/ (_| |\ V /| | (_| | || __/
// \_/\_/ \___|\__,_| \_/ |_|\__,_|\__\___|
//
// Copyright © 2016 - 2024 Weaviate B.V. All rights reserved.
//
// CONTACT: [email protected]
//
package classification
import (
"fmt"
"math"
"sort"
"strings"
)
// warning, not thread-safe for this spike
type TfIdfCalculator struct {
size int
documents []string
documentLengths []uint
docPointer int
terms map[string][]uint16
termIdf map[string]float32
}
func NewTfIdfCalculator(size int) *TfIdfCalculator {
return &TfIdfCalculator{
size: size,
documents: make([]string, size),
documentLengths: make([]uint, size),
terms: make(map[string][]uint16),
termIdf: make(map[string]float32),
}
}
func (c *TfIdfCalculator) AddDoc(doc string) error {
if c.docPointer > c.size {
return fmt.Errorf("doc size exceeded")
}
c.documents[c.docPointer] = doc
c.docPointer++
return nil
}
func (c *TfIdfCalculator) Calculate() {
for i := range c.documents {
c.analyzeDoc(i)
}
for term, frequencies := range c.terms {
var contained uint
for _, frequency := range frequencies {
if frequency > 0 {
contained++
}
}
c.termIdf[term] = float32(math.Log10(float64(c.size) / float64(contained)))
}
}
func (c *TfIdfCalculator) analyzeDoc(docIndex int) {
terms := newSplitter().Split(c.documents[docIndex])
for i, term := range terms {
term = strings.ToLower(term)
frequencies := c.getOrInitTerm(term)
frequencies[docIndex] = frequencies[docIndex] + 1
c.documentLengths[docIndex] = uint(i + 1)
c.terms[term] = frequencies
}
}
func (c *TfIdfCalculator) getOrInitTerm(term string) []uint16 {
frequencies, ok := c.terms[term]
if !ok {
frequencies := make([]uint16, c.size)
c.terms[term] = frequencies
return frequencies
}
return frequencies
}
func (c *TfIdfCalculator) Get(term string, doc int) float32 {
term = strings.ToLower(term)
frequencies, ok := c.terms[term]
if !ok {
return 0
}
tf := float32(frequencies[doc]) / float32(c.documentLengths[doc])
idf := c.termIdf[term]
return tf * idf
}
func (c *TfIdfCalculator) GetAllTerms(docIndex int) []TermWithTfIdf {
terms := newSplitter().Split(c.documents[docIndex])
terms = c.lowerCaseAndDedup(terms)
out := make([]TermWithTfIdf, len(terms))
for i, term := range terms {
out[i] = TermWithTfIdf{
Term: term,
TfIdf: c.Get(term, docIndex),
}
}
sort.Slice(out, func(a, b int) bool { return out[a].TfIdf > out[b].TfIdf })
return c.withRelativeScores(out)
}
type TermWithTfIdf struct {
Term string
TfIdf float32
RelativeScore float32
}
func (c *TfIdfCalculator) withRelativeScores(list []TermWithTfIdf) []TermWithTfIdf {
// mean for variance
var mean float64
for _, t := range list {
mean += float64(t.TfIdf)
}
mean = mean / float64(len(list))
// calculate variance
for i, t := range list {
variance := math.Pow(float64(t.TfIdf)-mean, 2)
if float64(t.TfIdf) < mean {
list[i].RelativeScore = float32(-variance)
} else {
list[i].RelativeScore = float32(variance)
}
}
return c.withNormalizedScores(list)
}
// between -1 and 1
func (c *TfIdfCalculator) withNormalizedScores(list []TermWithTfIdf) []TermWithTfIdf {
max, min := c.maxMin(list)
for i, curr := range list {
score := (curr.RelativeScore - min) / (max - min)
list[i].RelativeScore = (score - 0.5) * 2
}
return list
}
func (c *TfIdfCalculator) maxMin(list []TermWithTfIdf) (float32, float32) {
max := list[0].RelativeScore
min := list[0].RelativeScore
for _, curr := range list {
if curr.RelativeScore > max {
max = curr.RelativeScore
}
if curr.RelativeScore < min {
min = curr.RelativeScore
}
}
return max, min
}
func (c *TfIdfCalculator) lowerCaseAndDedup(list []string) []string {
seen := map[string]struct{}{}
out := make([]string, len(list))
i := 0
for _, term := range list {
term = strings.ToLower(term)
_, ok := seen[term]
if ok {
continue
}
seen[term] = struct{}{}
out[i] = term
i++
}
return out[:i]
}
|