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
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
|
// Copyright Vespa.ai. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
#include "hitcollector.h"
#include <vespa/searchlib/fef/feature_resolver.h>
#include <vespa/searchlib/fef/utils.h>
#include <vespa/vespalib/util/stringfmt.h>
#include <algorithm>
#include <vespa/log/log.h>
LOG_SETUP(".searchvisitor.hitcollector");
using vespalib::FeatureSet;
using vespalib::FeatureValues;
using vdslib::SearchResult;
using FefUtils = search::fef::Utils;
namespace streaming {
HitCollector::Hit::Hit(const vsm::StorageDocument * doc, uint32_t docId, const MatchData & matchData,
double score, const void * sortData, size_t sortDataLen) :
_docid(docId),
_score(score),
_document(doc),
_matchData(),
_sortBlob(sortData, sortDataLen)
{
_matchData.reserve(matchData.getNumTermFields());
for (search::fef::TermFieldHandle handle = 0; handle < matchData.getNumTermFields(); ++handle) {
_matchData.emplace_back(*matchData.resolveTermField(handle));
}
}
HitCollector::Hit::~Hit() = default;
HitCollector::HitCollector(size_t wantedHits, bool use_sort_blob) :
_hits(),
_use_sort_blob(use_sort_blob),
_sortedByDocId(true)
{
_hits.reserve(wantedHits);
}
const vsm::Document &
HitCollector::getDocSum(const search::DocumentIdT & docId) const
{
for (const Hit & hit : _hits) {
if (docId == hit.getDocId()) {
return hit.getDocument();
}
}
throw std::runtime_error(vespalib::make_string("Could not look up document id %d", docId));
}
bool
HitCollector::addHit(const vsm::StorageDocument * doc, uint32_t docId, const MatchData & data, double score)
{
return addHit(Hit(doc, docId, data, score));
}
bool
HitCollector::addHit(const vsm::StorageDocument * doc, uint32_t docId, const MatchData & data,
double score, const void * sortData, size_t sortDataLen)
{
return addHit(Hit(doc, docId, data, score, sortData, sortDataLen));
}
void
HitCollector::sortByDocId()
{
if (!_sortedByDocId) {
std::sort(_hits.begin(), _hits.end()); // sort on docId
_sortedByDocId = true;
}
}
bool
HitCollector::addHitToHeap(const Hit & hit) const
{
// return true if the given hit is better than the current worst one.
return (_use_sort_blob)
? (hit.cmpSort(_hits[0]) < 0)
: (hit.cmpRank(_hits[0]) < 0);
}
void
HitCollector::make_heap() {
if (_use_sort_blob) {
std::make_heap(_hits.begin(), _hits.end(), Hit::SortComparator());
} else {
std::make_heap(_hits.begin(), _hits.end(), Hit::RankComparator());
}
}
void
HitCollector::pop_heap() {
if (_use_sort_blob) {
std::pop_heap(_hits.begin(), _hits.end(), Hit::SortComparator());
} else {
std::pop_heap(_hits.begin(), _hits.end(), Hit::RankComparator());
}
}
void
HitCollector::push_heap() {
if (_use_sort_blob) {
std::push_heap(_hits.begin(), _hits.end(), Hit::SortComparator());
} else {
std::push_heap(_hits.begin(), _hits.end(), Hit::RankComparator());
}
}
bool
HitCollector::addHit(Hit && hit)
{
bool amongTheBest(false);
size_t avail = (_hits.capacity() - _hits.size());
assert(_use_sort_blob != hit.getSortBlob().empty() );
if (avail > 1) {
// No heap yet.
_hits.emplace_back(std::move(hit));
amongTheBest = true;
} else if (_hits.capacity() == 0) {
// this happens when wantedHitCount = 0
// in this case we shall not put anything on the heap (which is empty)
} else if ( avail == 0 && addHitToHeap(hit)) { // already a heap
pop_heap();
_hits.back() = std::move(hit);
amongTheBest = true;
push_heap();
} else if (avail == 1) { // make a heap of the hit vector
_hits.emplace_back(std::move(hit));
amongTheBest = true;
make_heap();
_sortedByDocId = false; // the hit vector is no longer sorted by docId
}
return amongTheBest;
}
void
HitCollector::fillSearchResult(vdslib::SearchResult & searchResult, FeatureValues&& match_features)
{
sortByDocId();
size_t count = std::min(_hits.size(), searchResult.getWantedHitCount());
for (size_t i(0); i < count; i++) {
const Hit & hit = _hits[i];
vespalib::string documentId(hit.getDocument().docDoc().getId().toString());
search::DocumentIdT docId = hit.getDocId();
SearchResult::RankType rank = hit.getRankScore();
LOG(debug, "fillSearchResult: gDocId(%s), lDocId(%u), rank(%f)", documentId.c_str(), docId, (float)rank);
if (hit.getSortBlob().empty()) {
searchResult.addHit(docId, documentId.c_str(), rank);
} else {
searchResult.addHit(docId, documentId.c_str(), rank, hit.getSortBlob().c_str(), hit.getSortBlob().size());
}
}
searchResult.set_match_features(std::move(match_features));
}
void
HitCollector::fillSearchResult(vdslib::SearchResult & searchResult)
{
fillSearchResult(searchResult, FeatureValues());
}
FeatureSet::SP
HitCollector::getFeatureSet(IRankProgram &rankProgram,
const FeatureResolver &resolver,
const search::StringStringMap &feature_rename_map)
{
if (resolver.num_features() == 0 || _hits.empty()) {
return std::make_shared<FeatureSet>();
}
sortByDocId();
auto names = FefUtils::extract_feature_names(resolver, feature_rename_map);
FeatureSet::SP retval = std::make_shared<FeatureSet>(names, _hits.size());
for (const Hit & hit : _hits) {
uint32_t docId = hit.getDocId();
rankProgram.run(docId, hit.getMatchData());
auto * f = retval->getFeaturesByIndex(retval->addDocId(docId));
FefUtils::extract_feature_values(resolver, docId, f);
}
return retval;
}
FeatureValues
HitCollector::get_match_features(IRankProgram& rank_program,
const FeatureResolver& resolver,
const search::StringStringMap& feature_rename_map)
{
FeatureValues match_features;
if (resolver.num_features() == 0 || _hits.empty()) {
return match_features;
}
sortByDocId();
match_features.names = FefUtils::extract_feature_names(resolver, feature_rename_map);
match_features.values.resize(resolver.num_features() * _hits.size());
auto f = match_features.values.data();
for (const Hit & hit : _hits) {
auto docid = hit.getDocId();
rank_program.run(docid, hit.getMatchData());
FefUtils::extract_feature_values(resolver, docid, f);
f += resolver.num_features();
}
assert(f == match_features.values.data() + match_features.values.size());
return match_features;
}
} // namespace streaming
|
