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
|
// Copyright 2019 Oath Inc. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
#include "emptysearch.h"
#include "nearest_neighbor_blueprint.h"
#include "nearest_neighbor_iterator.h"
#include "nns_index_iterator.h"
#include <vespa/searchlib/fef/termfieldmatchdataarray.h>
#include <vespa/eval/tensor/dense/dense_tensor_view.h>
#include <vespa/eval/tensor/dense/dense_tensor.h>
#include <vespa/searchlib/tensor/dense_tensor_attribute.h>
#include <vespa/searchlib/tensor/distance_function_factory.h>
using vespalib::tensor::DenseTensorView;
using vespalib::tensor::DenseTensor;
namespace search::queryeval {
namespace {
template<typename LCT, typename RCT>
void
convert_cells(std::unique_ptr<DenseTensorView> &original, vespalib::eval::ValueType want_type)
{
auto old_cells = original->cellsRef().typify<LCT>();
std::vector<RCT> new_cells;
new_cells.reserve(old_cells.size());
for (LCT value : old_cells) {
RCT conv = value;
new_cells.push_back(conv);
}
original = std::make_unique<DenseTensor<RCT>>(want_type, std::move(new_cells));
}
template<>
void
convert_cells<float,float>(std::unique_ptr<DenseTensorView> &, vespalib::eval::ValueType) {}
template<>
void
convert_cells<double,double>(std::unique_ptr<DenseTensorView> &, vespalib::eval::ValueType) {}
struct ConvertCellsSelector
{
template <typename LCT, typename RCT>
static auto get_fun() { return convert_cells<LCT, RCT>; }
};
} // namespace <unnamed>
NearestNeighborBlueprint::NearestNeighborBlueprint(const queryeval::FieldSpec& field,
const tensor::DenseTensorAttribute& attr_tensor,
std::unique_ptr<vespalib::tensor::DenseTensorView> query_tensor,
uint32_t target_num_hits, bool approximate, uint32_t explore_additional_hits)
: ComplexLeafBlueprint(field),
_attr_tensor(attr_tensor),
_query_tensor(std::move(query_tensor)),
_target_num_hits(target_num_hits),
_approximate(approximate),
_explore_additional_hits(explore_additional_hits),
_fallback_dist_fun(),
_distance_heap(target_num_hits),
_found_hits()
{
auto lct = _query_tensor->cellsRef().type;
auto rct = _attr_tensor.getTensorType().cell_type();
auto fixup_fun = vespalib::tensor::select_2<ConvertCellsSelector>(lct, rct);
fixup_fun(_query_tensor, _attr_tensor.getTensorType());
auto def_dm = search::attribute::DistanceMetric::Euclidean;
_fallback_dist_fun = search::tensor::make_distance_function(def_dm, rct);
_dist_fun = _fallback_dist_fun.get();
auto nns_index = _attr_tensor.nearest_neighbor_index();
if (nns_index) {
_dist_fun = nns_index->distance_function();
}
uint32_t est_hits = _attr_tensor.getNumDocs();
if (_approximate && nns_index) {
est_hits = std::min(target_num_hits, est_hits);
}
setEstimate(HitEstimate(est_hits, false));
set_want_global_filter(true);
}
NearestNeighborBlueprint::~NearestNeighborBlueprint() = default;
void
NearestNeighborBlueprint::set_global_filter(std::shared_ptr<BitVector> global_filter)
{
// XXX do something with global_filter here
(void) global_filter;
}
void
NearestNeighborBlueprint::perform_top_k()
{
auto nns_index = _attr_tensor.nearest_neighbor_index();
if (_approximate && nns_index) {
auto lhs_type = _query_tensor->fast_type();
auto rhs_type = _attr_tensor.getTensorType();
// XXX deal with different cell types later
if (lhs_type == rhs_type) {
auto lhs = _query_tensor->cellsRef();
uint32_t k = _target_num_hits;
_found_hits = nns_index->find_top_k(k, lhs, k + _explore_additional_hits);
}
}
}
void
NearestNeighborBlueprint::fetchPostings(const ExecuteInfo &execInfo) {
if (execInfo.isStrict()) {
perform_top_k();
}
}
std::unique_ptr<SearchIterator>
NearestNeighborBlueprint::createLeafSearch(const search::fef::TermFieldMatchDataArray& tfmda, bool strict) const
{
assert(tfmda.size() == 1);
fef::TermFieldMatchData &tfmd = *tfmda[0]; // always search in only one field
if (strict && ! _found_hits.empty()) {
return NnsIndexIterator::create(tfmd, _found_hits, _dist_fun);
}
const vespalib::tensor::DenseTensorView &qT = *_query_tensor;
return NearestNeighborIterator::create(strict, tfmd, qT, _attr_tensor, _distance_heap, _dist_fun);
}
void
NearestNeighborBlueprint::visitMembers(vespalib::ObjectVisitor& visitor) const
{
ComplexLeafBlueprint::visitMembers(visitor);
visitor.visitString("attribute_tensor", _attr_tensor.getTensorType().to_spec());
visitor.visitString("query_tensor", _query_tensor->type().to_spec());
visitor.visitInt("target_num_hits", _target_num_hits);
visitor.visitBool("approximate", _approximate);
visitor.visitInt("explore_additional_hits", _explore_additional_hits);
}
bool
NearestNeighborBlueprint::always_needs_unpack() const
{
return true;
}
}
|
