Spaces:
Running
Running
File size: 3,583 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 |
// _ _
// __ _____ __ ___ ___ __ _| |_ ___
// \ \ /\ / / _ \/ _` \ \ / / |/ _` | __/ _ \
// \ V V / __/ (_| |\ V /| | (_| | || __/
// \_/\_/ \___|\__,_| \_/ |_|\__,_|\__\___|
//
// Copyright © 2016 - 2024 Weaviate B.V. All rights reserved.
//
// CONTACT: [email protected]
//
package grouper
import (
"fmt"
"github.com/sirupsen/logrus"
"github.com/weaviate/weaviate/entities/search"
"github.com/weaviate/weaviate/usecases/vectorizer"
)
// Grouper groups or merges search results by how related they are
type Grouper struct {
logger logrus.FieldLogger
}
// NewGrouper creates a Grouper UC from the specified configuration
func New(logger logrus.FieldLogger) *Grouper {
return &Grouper{logger: logger}
}
// Group using the applied strategy and force
func (g *Grouper) Group(in []search.Result, strategy string,
force float32,
) ([]search.Result, error) {
groups := groups{logger: g.logger}
for _, current := range in {
pos, ok := groups.hasMatch(current.Vector, force)
if !ok {
groups.new(current)
} else {
groups.Elements[pos].add(current)
}
}
return groups.flatten(strategy)
}
type group struct {
Elements []search.Result `json:"elements"`
}
func (g *group) add(item search.Result) {
g.Elements = append(g.Elements, item)
}
func (g group) matches(vector []float32, force float32) bool {
// iterate over all group Elements and consider it a match if any matches
for _, elem := range g.Elements {
dist, err := vectorizer.NormalizedDistance(vector, elem.Vector)
if err != nil {
// TODO: log error
// we don't expect to ever see this error, so we don't need to handle it
// explicitly, however, let's still log it in case that the above
// assumption is wrong
continue
}
if dist < force {
return true
}
}
return false
}
type groups struct {
Elements []group `json:"elements"`
logger logrus.FieldLogger
}
func (gs groups) hasMatch(vector []float32, force float32) (int, bool) {
for pos, group := range gs.Elements {
if group.matches(vector, force) {
return pos, true
}
}
return -1, false
}
func (gs *groups) new(item search.Result) {
gs.Elements = append(gs.Elements, group{Elements: []search.Result{item}})
}
func (gs groups) flatten(strategy string) (out []search.Result, err error) {
gs.logger.WithField("object", "grouping_before_flatten").
WithField("strategy", strategy).
WithField("groups", gs.Elements).
Debug("group before flattening")
switch strategy {
case "closest":
out, err = gs.flattenClosest()
case "merge":
out, err = gs.flattenMerge()
default:
return nil, fmt.Errorf("unrecognized grouping strategy '%s'", strategy)
}
if err != nil {
return
}
gs.logger.WithField("object", "grouping_after_flatten").
WithField("strategy", strategy).
WithField("groups", gs.Elements).
Debug("group after flattening")
return out, nil
}
func (gs groups) flattenClosest() ([]search.Result, error) {
out := make([]search.Result, len(gs.Elements))
for i, group := range gs.Elements {
out[i] = group.Elements[0] // hard-code "closest" strategy for now
}
return out, nil
}
func (gs groups) flattenMerge() ([]search.Result, error) {
out := make([]search.Result, len(gs.Elements))
for i, group := range gs.Elements {
merged, err := group.flattenMerge()
if err != nil {
return nil, fmt.Errorf("group %d: %v", i, err)
}
out[i] = merged
}
return out, nil
}
|