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
|
// Copyright Yahoo. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
#include <vespa/document/base/fieldpath.h>
#include <vespa/document/datatype/tensor_data_type.h>
#include <vespa/document/fieldvalue/tensorfieldvalue.h>
#include <vespa/eval/eval/simple_value.h>
#include <vespa/eval/eval/tensor_spec.h>
#include <vespa/eval/eval/value_codec.h>
#include <vespa/eval/eval/value_type.h>
#include <vespa/searchlib/fef/indexproperties.h>
#include <vespa/searchlib/query/streaming/nearest_neighbor_query_node.h>
#include <vespa/searchlib/tensor/euclidean_distance.h>
#include <vespa/vespalib/gtest/gtest.h>
#include <vespa/vsm/searcher/mock_field_searcher_env.h>
#include <vespa/vsm/searcher/nearest_neighbor_field_searcher.h>
using namespace search::attribute::test;
using namespace search::attribute;
using namespace search::fef;
using namespace search::streaming;
using namespace search::tensor;
using namespace vespalib::eval;
using namespace vsm;
using document::FieldPath;
using document::FieldPathEntry;
using document::TensorDataType;
using document::TensorFieldValue;
struct MockQuery {
std::vector<std::unique_ptr<NearestNeighborQueryNode>> nodes;
QueryTermList term_list;
MockQuery& add(const vespalib::string& query_tensor_name,
uint32_t target_hits,
double distance_threshold) {
std::unique_ptr<QueryNodeResultBase> base;
auto node = std::make_unique<NearestNeighborQueryNode>(std::move(base), query_tensor_name, "my_tensor_field",
target_hits, distance_threshold, 7, search::query::Weight(100));
nodes.push_back(std::move(node));
term_list.push_back(nodes.back().get());
return *this;
}
~MockQuery() {}
const NearestNeighborQueryNode& get(size_t idx) const {
assert(idx < nodes.size());
return *nodes[idx];
}
void reset() {
for (auto term : term_list) {
term->reset();
}
}
};
class NearestNeighborSearcherTest : public testing::Test {
public:
vsm::test::MockFieldSearcherEnv env;
ValueType tensor_type;
TensorDataType data_type;
vsm::FieldIdT field_id;
NearestNeighborFieldSearcher searcher;
MockQuery query;
NearestNeighborSearcherTest()
: env(),
tensor_type(ValueType::from_spec("tensor(x[2])")),
data_type(tensor_type),
field_id(2),
searcher(field_id, DistanceMetric::Euclidean),
query()
{
env.field_paths->resize(field_id + 1);
(*env.field_paths)[field_id].push_back(std::make_unique<FieldPathEntry>(data_type, "my_tensor_field"));
}
void set_query_tensor(const vespalib::string& query_tensor_name,
const vespalib::string& spec_expr) {
search::fef::indexproperties::type::QueryFeature::set(env.index_env.getProperties(), query_tensor_name, tensor_type.to_spec());
auto tensor = SimpleValue::from_spec(TensorSpec::from_expr(spec_expr));
vespalib::nbostream stream;
vespalib::eval::encode_value(*tensor, stream);
env.query_props.add(query_tensor_name, vespalib::stringref(stream.peek(), stream.size()));
}
void prepare() {
env.prepare(searcher, query.term_list);
}
void match(const vespalib::string& spec_expr) {
TensorFieldValue fv(data_type);
auto tensor = SimpleValue::from_spec(TensorSpec::from_expr(spec_expr));
fv = std::move(tensor);
query.reset();
searcher.onValue(fv);
}
void expect_match(const vespalib::string& spec_expr, double exp_square_distance, const NearestNeighborQueryNode& node) {
match(spec_expr);
expect_match(exp_square_distance, node);
}
void expect_match(double exp_square_distance, const NearestNeighborQueryNode& node) {
double exp_raw_score = 1.0 / (1.0 + std::sqrt(exp_square_distance));
EXPECT_TRUE(node.evaluate());
EXPECT_DOUBLE_EQ(exp_square_distance, node.get_distance().value());
EXPECT_DOUBLE_EQ(exp_raw_score, node.get_raw_score().value());
}
void expect_not_match(const vespalib::string& spec_expr, const NearestNeighborQueryNode& node) {
match(spec_expr);
EXPECT_FALSE(node.evaluate());
}
};
TEST_F(NearestNeighborSearcherTest, distance_heap_keeps_the_best_target_hits)
{
query.add("qt1", 2, 100.0);
const auto& node = query.get(0);
set_query_tensor("qt1", "tensor(x[2]):[1,3]");
prepare();
expect_match("tensor(x[2]):[1,7]", (7-3)*(7-3), node);
expect_match("tensor(x[2]):[1,9]", (9-3)*(9-3), node);
// The distance limit is now (9-3)*(9-3) = 36, so this is not good enough.
expect_not_match("tensor(x[2]):[1,10]", node);
expect_match("tensor(x[2]):[1,5]", (5-3)*(5-3), node);
// The distance limit is now (7-3)*(7-3) = 16, so this is not good enough.
expect_not_match("tensor(x[2]):[1,8]", node);
// This is not considered a document match as get_raw_score() is not called,
// and the distance heap is not updated.
match("tensor(x[2]):[1,4]");
EXPECT_EQ(1, node.get_distance().value());
EXPECT_TRUE(node.evaluate());
// The distance limit is still (7-3)*(7-3) = 16, so this is in fact good enough.
expect_match("tensor(x[2]):[1,6]", (6-3)*(6-3), node);
// The distance limit is (6-3)*(6-3) = 4, and a similar distance is a match.
expect_match("tensor(x[2]):[1,6]", (6-3)*(6-3), node);
}
TEST_F(NearestNeighborSearcherTest, raw_score_calculated_with_distance_threshold)
{
query.add("qt1", 10, 3.0);
const auto& node = query.get(0);
set_query_tensor("qt1", "tensor(x[2]):[1,3]");
prepare();
expect_match("tensor(x[2]):[1,5]", (5-3)*(5-3), node);
expect_match("tensor(x[2]):[1,6]", (6-3)*(6-3), node);
// This is not a match since ((7-3)*(7-3) = 16) is larger than the internal distance threshold of (3*3 = 9).
expect_not_match("tensor(x[2]):[1,7]", node);
}
TEST_F(NearestNeighborSearcherTest, raw_score_calculated_for_two_query_operators)
{
query.add("qt1", 10, 3.0);
query.add("qt2", 10, 4.0);
set_query_tensor("qt1", "tensor(x[2]):[1,3]");
set_query_tensor("qt2", "tensor(x[2]):[1,4]");
prepare();
match("tensor(x[2]):[1,5]");
expect_match((5-3)*(5-3), query.get(0));
expect_match((5-4)*(5-4), query.get(1));
match("tensor(x[2]):[1,7]");
// This is not a match since ((7-3)*(7-3) = 16) is larger than the internal distance threshold of (3*3 = 9).
EXPECT_FALSE(query.get(0).evaluate());
expect_match((7-4)*(7-4), query.get(1));
}
TEST_F(NearestNeighborSearcherTest, distance_metric_from_string)
{
using NNFS = NearestNeighborFieldSearcher;
EXPECT_EQ(DistanceMetric::Euclidean, NNFS::distance_metric_from_string("EUCLIDEAN"));
EXPECT_EQ(DistanceMetric::Angular, NNFS::distance_metric_from_string("ANGULAR"));
EXPECT_EQ(DistanceMetric::GeoDegrees, NNFS::distance_metric_from_string("GEODEGREES"));
EXPECT_EQ(DistanceMetric::InnerProduct, NNFS::distance_metric_from_string("INNERPRODUCT"));
EXPECT_EQ(DistanceMetric::Hamming, NNFS::distance_metric_from_string("HAMMING"));
EXPECT_EQ(DistanceMetric::Euclidean, NNFS::distance_metric_from_string("not_available"));
}
GTEST_MAIN_RUN_ALL_TESTS()
|
