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
|
// Copyright Yahoo. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
#include <vespa/vespalib/stllike/string.h>
#include <vespa/vespalib/util/stringfmt.h>
#include <vespa/vespalib/util/issue.h>
#include <vespa/searchlib/features/rankingexpressionfeature.h>
#include <vespa/searchlib/features/onnx_feature.h>
#include <vespa/searchlib/fef/blueprintfactory.h>
#include <vespa/searchlib/fef/indexproperties.h>
#include <vespa/searchlib/fef/matchdatalayout.h>
#include <vespa/searchlib/fef/test/indexenvironment.h>
#include <vespa/searchlib/fef/test/queryenvironment.h>
#include <vespa/searchlib/fef/rank_program.h>
#include <vespa/searchlib/fef/test/test_features.h>
#include <vespa/vespalib/gtest/gtest.h>
using namespace search::fef;
using namespace search::fef::test;
using namespace search::features;
using vespalib::make_string_short::fmt;
using vespalib::eval::TensorSpec;
using vespalib::Issue;
std::string get_source_dir() {
const char *dir = getenv("SOURCE_DIRECTORY");
return (dir ? dir : ".");
}
std::string source_dir = get_source_dir();
std::string vespa_dir = source_dir + "/" + "../../../../..";
std::string simple_model = vespa_dir + "/" + "eval/src/tests/tensor/onnx_wrapper/simple.onnx";
std::string dynamic_model = vespa_dir + "/" + "eval/src/tests/tensor/onnx_wrapper/dynamic.onnx";
std::string strange_names_model = source_dir + "/" + "strange_names.onnx";
std::string fragile_model = source_dir + "/" + "fragile.onnx";
uint32_t default_docid = 1;
vespalib::string expr_feature(const vespalib::string &name) {
return fmt("rankingExpression(%s)", name.c_str());
}
vespalib::string onnx_feature(const vespalib::string &name) {
return fmt("onnx(%s)", name.c_str());
}
vespalib::string onnx_feature_old(const vespalib::string &name) {
return fmt("onnxModel(%s)", name.c_str());
}
struct OnnxFeatureTest : ::testing::Test {
BlueprintFactory factory;
IndexEnvironment indexEnv;
BlueprintResolver::SP resolver;
Properties overrides;
MatchData::UP match_data;
RankProgram program;
OnnxFeatureTest() : factory(), indexEnv(), resolver(new BlueprintResolver(factory, indexEnv)),
overrides(), match_data(), program(resolver)
{
factory.addPrototype(std::make_shared<DocidBlueprint>());
factory.addPrototype(std::make_shared<RankingExpressionBlueprint>());
factory.addPrototype(std::make_shared<OnnxBlueprint>("onnx"));
factory.addPrototype(std::make_shared<OnnxBlueprint>("onnxModel"));
}
~OnnxFeatureTest() override;
void add_expr(const vespalib::string &name, const vespalib::string &expr) {
vespalib::string feature_name = expr_feature(name);
vespalib::string expr_name = feature_name + ".rankingScript";
indexEnv.getProperties().add(expr_name, expr);
}
void add_onnx(OnnxModel model) {
indexEnv.addOnnxModel(std::move(model));
}
bool try_compile(const vespalib::string &seed) {
resolver->addSeed(seed);
if (!resolver->compile()) {
return false;
}
MatchDataLayout mdl;
QueryEnvironment queryEnv(&indexEnv);
match_data = mdl.createMatchData();
program.setup(*match_data, queryEnv, overrides);
return true;
}
void compile(const vespalib::string &seed) {
ASSERT_TRUE(try_compile(seed));
}
TensorSpec get(const vespalib::string &feature, uint32_t docid) const {
auto result = program.get_all_features(false);
for (size_t i = 0; i < result.num_features(); ++i) {
if (result.name_of(i) == feature) {
return TensorSpec::from_value(result.resolve(i).as_object(docid));
}
}
return {"error"};
}
TensorSpec get(uint32_t docid) const {
auto result = program.get_seeds(false);
EXPECT_EQ(1u, result.num_features());
return TensorSpec::from_value(result.resolve(0).as_object(docid));
}
};
OnnxFeatureTest::~OnnxFeatureTest() = default;
TEST_F(OnnxFeatureTest, simple_onnx_model_can_be_calculated) {
add_expr("query_tensor", "tensor<float>(a[1],b[4]):[[docid,2,3,4]]");
add_expr("attribute_tensor", "tensor<float>(a[4],b[1]):[[5],[6],[7],[8]]");
add_expr("bias_tensor", "tensor<float>(a[1],b[1]):[[9]]");
add_onnx(OnnxModel("simple", simple_model));
compile(onnx_feature("simple"));
EXPECT_EQ(get(1), TensorSpec("tensor<float>(d0[1],d1[1])").add({{"d0",0},{"d1",0}}, 79.0));
EXPECT_EQ(get("onnx(simple).output", 1), TensorSpec("tensor<float>(d0[1],d1[1])").add({{"d0",0},{"d1",0}}, 79.0));
EXPECT_EQ(get(2), TensorSpec("tensor<float>(d0[1],d1[1])").add({{"d0",0},{"d1",0}}, 84.0));
EXPECT_EQ(get(3), TensorSpec("tensor<float>(d0[1],d1[1])").add({{"d0",0},{"d1",0}}, 89.0));
}
TEST_F(OnnxFeatureTest, simple_onnx_model_can_be_calculated_with_old_name) {
add_expr("query_tensor", "tensor<float>(a[1],b[4]):[[docid,2,3,4]]");
add_expr("attribute_tensor", "tensor<float>(a[4],b[1]):[[5],[6],[7],[8]]");
add_expr("bias_tensor", "tensor<float>(a[1],b[1]):[[9]]");
add_onnx(OnnxModel("simple", simple_model));
compile(onnx_feature_old("simple"));
EXPECT_EQ(get(1), TensorSpec("tensor<float>(d0[1],d1[1])").add({{"d0",0},{"d1",0}}, 79.0));
EXPECT_EQ(get("onnxModel(simple).output", 1), TensorSpec("tensor<float>(d0[1],d1[1])").add({{"d0",0},{"d1",0}}, 79.0));
EXPECT_EQ(get(2), TensorSpec("tensor<float>(d0[1],d1[1])").add({{"d0",0},{"d1",0}}, 84.0));
EXPECT_EQ(get(3), TensorSpec("tensor<float>(d0[1],d1[1])").add({{"d0",0},{"d1",0}}, 89.0));
}
TEST_F(OnnxFeatureTest, dynamic_onnx_model_can_be_calculated) {
add_expr("query_tensor", "tensor<float>(a[1],b[4]):[[docid,2,3,4]]");
add_expr("attribute_tensor", "tensor<float>(a[4],b[1]):[[5],[6],[7],[8]]");
add_expr("bias_tensor", "tensor<float>(a[1],b[2]):[[4,5]]");
add_onnx(OnnxModel("dynamic", dynamic_model));
compile(onnx_feature("dynamic"));
EXPECT_EQ(get(1), TensorSpec("tensor<float>(d0[1],d1[1])").add({{"d0",0},{"d1",0}}, 79.0));
EXPECT_EQ(get("onnx(dynamic).output", 1), TensorSpec("tensor<float>(d0[1],d1[1])").add({{"d0",0},{"d1",0}}, 79.0));
EXPECT_EQ(get(2), TensorSpec("tensor<float>(d0[1],d1[1])").add({{"d0",0},{"d1",0}}, 84.0));
EXPECT_EQ(get(3), TensorSpec("tensor<float>(d0[1],d1[1])").add({{"d0",0},{"d1",0}}, 89.0));
}
TEST_F(OnnxFeatureTest, strange_input_and_output_names_are_normalized) {
add_expr("input_0", "tensor<float>(a[2]):[10,20]");
add_expr("input_1", "tensor<float>(a[2]):[5,10]");
add_onnx(OnnxModel("strange_names", strange_names_model));
compile(onnx_feature("strange_names"));
auto expect_add = TensorSpec("tensor<float>(d0[2])").add({{"d0",0}},15).add({{"d0",1}},30);
auto expect_sub = TensorSpec("tensor<float>(d0[2])").add({{"d0",0}},5).add({{"d0",1}},10);
EXPECT_EQ(get(1), expect_add);
EXPECT_EQ(get("onnx(strange_names).foo_bar", 1), expect_add);
EXPECT_EQ(get("onnx(strange_names)._baz_0", 1), expect_sub);
}
TEST_F(OnnxFeatureTest, input_features_and_output_names_can_be_specified) {
add_expr("my_first_input", "tensor<float>(a[2]):[10,20]");
add_expr("my_second_input", "tensor<float>(a[2]):[5,10]");
add_onnx(std::move(OnnxModel("custom_names", strange_names_model)
.input_feature("input:0", "rankingExpression(my_first_input)")
.input_feature("input/1", "rankingExpression(my_second_input)")
.output_name("foo/bar", "my_first_output")
.output_name("-baz:0", "my_second_output")));
compile(onnx_feature("custom_names"));
auto expect_add = TensorSpec("tensor<float>(d0[2])").add({{"d0",0}},15).add({{"d0",1}},30);
auto expect_sub = TensorSpec("tensor<float>(d0[2])").add({{"d0",0}},5).add({{"d0",1}},10);
EXPECT_EQ(get(1), expect_add);
EXPECT_EQ(get("onnx(custom_names).my_first_output", 1), expect_add);
EXPECT_EQ(get("onnx(custom_names).my_second_output", 1), expect_sub);
}
TEST_F(OnnxFeatureTest, fragile_model_can_be_evaluated) {
add_expr("in1", "tensor<float>(x[2]):[docid,5]");
add_expr("in2", "tensor<float>(x[2]):[docid,10]");
add_onnx(std::move(OnnxModel("fragile", fragile_model).dry_run_on_setup(true)));
EXPECT_TRUE(try_compile(onnx_feature("fragile")));
EXPECT_EQ(get(1), TensorSpec::from_expr("tensor<float>(d0[2]):[2,15]"));
EXPECT_EQ(get(3), TensorSpec::from_expr("tensor<float>(d0[2]):[6,15]"));
}
struct MyIssues : Issue::Handler {
std::vector<vespalib::string> list;
Issue::Binding capture;
MyIssues() : list(), capture(Issue::listen(*this)) {}
~MyIssues() override;
void handle(const Issue &issue) override { list.push_back(issue.message()); }
};
MyIssues::~MyIssues() = default;
TEST_F(OnnxFeatureTest, broken_model_evaluates_to_all_zeros) {
add_expr("in1", "tensor<float>(x[2]):[docid,5]");
add_expr("in2", "tensor<float>(x[3]):[docid,10,31515]");
add_onnx(std::move(OnnxModel("fragile", fragile_model).dry_run_on_setup(false)));
EXPECT_TRUE(try_compile(onnx_feature("fragile")));
MyIssues my_issues;
EXPECT_EQ(my_issues.list.size(), 0);
EXPECT_EQ(get(1), TensorSpec::from_expr("tensor<float>(d0[2]):[0,0]"));
EXPECT_EQ(my_issues.list.size(), 1);
EXPECT_EQ(get(3), TensorSpec::from_expr("tensor<float>(d0[2]):[0,0]"));
ASSERT_EQ(my_issues.list.size(), 2);
EXPECT_EQ(my_issues.list[0], my_issues.list[1]);
}
TEST_F(OnnxFeatureTest, broken_model_fails_with_dry_run) {
add_expr("in1", "tensor<float>(x[2]):[docid,5]");
add_expr("in2", "tensor<float>(x[3]):[docid,10,31515]");
add_onnx(std::move(OnnxModel("fragile", fragile_model).dry_run_on_setup(true)));
EXPECT_FALSE(try_compile(onnx_feature("fragile")));
}
GTEST_MAIN_RUN_ALL_TESTS()
|
