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
|
// Copyright Yahoo. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
#include "nearest_neighbor_field_searcher.h"
#include <vespa/document/datatype/datatype.h>
#include <vespa/document/datatype/tensor_data_type.h>
#include <vespa/document/fieldvalue/tensorfieldvalue.h>
#include <vespa/searchcommon/attribute/config.h>
#include <vespa/searchlib/fef/iqueryenvironment.h>
#include <vespa/searchlib/fef/query_value.h>
#include <vespa/searchlib/query/streaming/nearest_neighbor_query_node.h>
#include <vespa/searchlib/tensor/distance_calculator.h>
#include <vespa/searchlib/tensor/distance_function.h>
#include <vespa/searchlib/tensor/distance_function_factory.h>
#include <vespa/searchlib/tensor/tensor_ext_attribute.h>
#include <vespa/vespalib/util/exceptions.h>
#include <vespa/vespalib/util/issue.h>
using search::attribute::BasicType;
using search::attribute::CollectionType;
using search::attribute::Config;
using search::fef::QueryValue;
using search::tensor::DistanceCalculator;
using search::tensor::TensorExtAttribute;
using search::tensor::make_distance_function;
using vespalib::eval::ValueType;
namespace {
constexpr uint32_t scratch_docid = 0;
std::unique_ptr<TensorExtAttribute>
make_attribute(const ValueType& tensor_type)
{
Config cfg(BasicType::TENSOR, CollectionType::SINGLE);
cfg.setTensorType(tensor_type);
auto result = std::make_unique<TensorExtAttribute>("nnfs_attr", cfg);
uint32_t docid;
result->addDoc(docid);
assert(docid == scratch_docid);
return result;
}
}
namespace vsm {
NearestNeighborFieldSearcher::NodeAndCalc::NodeAndCalc(search::streaming::NearestNeighborQueryNode* node_in,
std::unique_ptr<search::tensor::DistanceCalculator> calc_in)
: node(node_in),
calc(std::move(calc_in)),
distance_threshold(calc->function().convert_threshold(node->get_distance_threshold()))
{
}
NearestNeighborFieldSearcher::NearestNeighborFieldSearcher(const FieldIdT& fid,
search::attribute::DistanceMetric metric)
: FieldSearcher(fid),
_metric(metric),
_attr(),
_calcs()
{
}
NearestNeighborFieldSearcher::~NearestNeighborFieldSearcher() = default;
std::unique_ptr<FieldSearcher>
NearestNeighborFieldSearcher::duplicate() const
{
return std::make_unique<NearestNeighborFieldSearcher>(field(), _metric);
}
void
NearestNeighborFieldSearcher::prepare(search::streaming::QueryTermList& qtl,
const SharedSearcherBuf& buf,
const vsm::FieldPathMapT& field_paths,
search::fef::IQueryEnvironment& query_env)
{
FieldSearcher::prepare(qtl, buf, field_paths, query_env);
const auto* tensor_type = field_paths[field()].back().getDataType().cast_tensor();
if (tensor_type == nullptr) {
vespalib::Issue::report("Data type for field %u is '%s', but expected it to be a tensor type",
field(), field_paths[field()].back().getDataType().toString().c_str());
}
_attr = make_attribute(tensor_type->getTensorType());
_calcs.clear();
for (auto term : qtl) {
auto* nn_term = term->as_nearest_neighbor_query_node();
if (nn_term == nullptr) {
vespalib::Issue::report("Query term (%s) searching field %u is NOT a NearestNeighborQueryNode",
term->getClassName().c_str(), field());
continue;
}
auto query_value = QueryValue::from_config(nn_term->get_query_tensor_name(), query_env.getIndexEnvironment());
query_value.prepare_shared_state(query_env, query_env.getObjectStore());
const auto* tensor_value = query_value.lookup_value(query_env.getObjectStore());
if (tensor_value == nullptr) {
vespalib::Issue::report("Could not find query tensor for NearestNeighborQueryNode(%s, %s)",
nn_term->index().c_str(), nn_term->get_query_tensor_name().c_str());
continue;
}
try {
auto calc = DistanceCalculator::make_with_validation(*_attr, *tensor_value);
_calcs.emplace_back(nn_term, std::move(calc));
} catch (const vespalib::IllegalArgumentException& ex) {
vespalib::Issue::report("Could not create DistanceCalculator for NearestNeighborQueryNode(%s, %s): %s",
nn_term->index().c_str(), nn_term->get_query_tensor_name().c_str(), ex.what());
}
}
}
void
NearestNeighborFieldSearcher::onValue(const document::FieldValue& fv)
{
if (fv.isA(document::FieldValue::Type::TENSOR)) {
const auto* tfv = dynamic_cast<const document::TensorFieldValue*>(&fv);
if (tfv && tfv->getAsTensorPtr()) {
_attr->add(*tfv->getAsTensorPtr(), 1);
for (auto& elem : _calcs) {
double distance = elem.calc->calc_with_limit(scratch_docid, elem.distance_threshold);
if (distance <= elem.distance_threshold) {
double score = elem.calc->function().to_rawscore(distance);
elem.node->set_raw_score(score);
}
}
}
}
}
search::attribute::DistanceMetric
NearestNeighborFieldSearcher::distance_metric_from_string(const vespalib::string& value)
{
using search::attribute::DistanceMetric;
// Valid string values must match the definition of DistanceMetric in
// config-model/src/main/java/com/yahoo/schema/document/Attribute.java
if (value == "EUCLIDEAN") {
return DistanceMetric::Euclidean;
} else if (value == "ANGULAR") {
return DistanceMetric::Angular;
} else if (value == "GEODEGREES") {
return DistanceMetric::GeoDegrees;
} else if (value == "INNERPRODUCT") {
return DistanceMetric::InnerProduct;
} else if (value == "HAMMING") {
return DistanceMetric::Hamming;
}
vespalib::Issue::report("Distance metric '%s' is not supported. Using 'euclidean' instead", value.c_str());
return DistanceMetric::Euclidean;
}
}
|
