proximity-cut/test.cpp at master · IUCVLab/proximity-cut · GitHub

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

#include <ctime>
#include <fstream>
#include <queue>
#include <chrono>
#include <utility>

#include "test_tools.h"


void run_test(const size_t count, const size_t vecdim, const size_t M) {
	std::cout << "|V|=" << count << "; Dim=" << vecdim << "; M=" << M << ". Starting index building. " << std::flush;
	hnswlib::L2Space l2space(vecdim);
	float *data, *testdata;
	hnswlib::labeltype *classes, *testclasses;
	hnswlib::HierarchicalNSW<float> index = hnswlib::HierarchicalNSW<float>(&l2space, count * 10, M);
	size_t pos_cls = 0;
	auto generated = generate_dataset(vecdim, count);
	data = generated.first;
	classes = generated.second;
	for (size_t i = 0; i < count; ++i) {
		pos_cls += 1 == classes[i];
		index.addPoint((void*)(data + i * vecdim), i, classes[i]);
	}

	std::cout << "Index is created. " << pos_cls << " * '1' / " << count << " total. Starting a test" << std::endl;

	size_t testsize = count / 20,
					positive_path = 0,
					positive_nsw_path = 0,
					positive_1nn = 0,
				  positive_5nn = 0;
	float nsw_time = 0, knn_time = 0, knn5_time = 0, path_time = 0;

	auto generated_test = generate_dataset(vecdim, testsize);
	testdata = generated_test.first;
	testclasses = generated_test.second;

	{
		auto start = std::chrono::high_resolution_clock::now();
		for (size_t i = 0; i < testsize; ++i) {
			positive_1nn += index.getPointClass(index.searchKnn((void*)(testdata + i * vecdim), 1).top().second) == testclasses[i];
		}
		auto stop = std::chrono::high_resolution_clock::now();
		knn_time = std::chrono::duration_cast<std::chrono::microseconds>(stop - start).count();
		std::cout << "     1-NN: " << 100. * positive_1nn / testsize << "% in " << knn_time / testsize << " ms avg" << std::endl;
	}

	{
		size_t k = 5;
		auto start = std::chrono::high_resolution_clock::now();
		for (size_t i = 0; i < testsize; ++i) {
			std::map<hnswlib::labeltype, size_t> m;
			auto q = index.searchKnn((void*)(testdata + i * vecdim), k);
			while (!q.empty()) {
				auto p = q.top();
				q.pop();
				m[index.getPointClass(p.second)]++;
			}
			hnswlib::labeltype best;
			size_t best_count = 0;
			for (const auto& pair : m) {
				if (pair.second > best_count) {
					best = pair.first;
					best_count = pair.second;
				}
			}
			positive_5nn += best == testclasses[i];
		}
		auto stop = std::chrono::high_resolution_clock::now();
		knn5_time = std::chrono::duration_cast<std::chrono::microseconds>(stop - start).count();
		std::cout << "     5-NN: " << 100. * positive_5nn / testsize << "% in " << knn5_time / testsize << " ms avg" << std::endl;
	}

	{
		auto start = std::chrono::high_resolution_clock::now();
		for (size_t i = 0; i < testsize; ++i) {
			positive_nsw_path += index.classifyByPathNSW((void*)(testdata + i * vecdim)) == testclasses[i];
		}
		auto stop = std::chrono::high_resolution_clock::now();
		nsw_time = std::chrono::duration_cast<std::chrono::microseconds>(stop - start).count();
		std::cout << " NSW-PATH: " << 100. * positive_nsw_path / testsize << "% in " << nsw_time / testsize << " ms avg" << std::endl;
	}

	{
		auto start = std::chrono::high_resolution_clock::now();
		for (size_t i = 0; i < testsize; ++i) {
			positive_path += index.classifyByPath((void*)(testdata + i * vecdim)) == testclasses[i];
		}
		auto stop = std::chrono::high_resolution_clock::now();
		path_time = std::chrono::duration_cast<std::chrono::microseconds>(stop - start).count();
		std::cout << "HNSW-PATH: " << 100. * positive_path / testsize << "% in " << path_time / testsize << " ms avg" << std::endl;
	}

	std::cout << " NSWvs1NN: " << "speedup " << knn_time / nsw_time << " times avg" << std::endl;
	std::cout << "HNSWvs1NN: " << "speedup " << knn_time / path_time << " times avg" << std::endl;
	std::cout << " NSWvs5NN: " << "speedup " << knn5_time / nsw_time << " times avg" << std::endl;
	std::cout << "HNSWvs5NN: " << "speedup " << knn5_time / path_time << " times avg" << std::endl;

	delete data,
	delete testdata;
	delete classes;
	delete testclasses;
}

int main() {
	srand(time(NULL));

	// std::cout << "'=== HIGGS 32 ===" << std::endl;
	// dataset_t higgs = load_libsvm("./data/higgs/HIGGS", 28, 11000000);
	// run_test_general(higgs, 28, 11000000, .99, 32);


	// std::cout << "'=== SUSY 64 ===" << std::endl;
	// dataset_t susy = load_libsvm("./data/susy/SUSY", 18, 5000000);
	// run_test_general(susy, 18, 5000000, .99, 64);

	dataset_t covertype = load_libsvm("./data/covtype/covtype.binary.scale", 54, 581012);
	run_test_stat(covertype, 54, 58101, .95, 4);
	covertype = load_libsvm("./data/covtype/covtype.binary.scale", 54, 581012);
	run_test_stat(covertype, 54, 58101, .95, 8);
	covertype = load_libsvm("./data/covtype/covtype.binary.scale", 54, 581012);
	run_test_stat(covertype, 54, 58101, .95, 16);
	covertype = load_libsvm("./data/covtype/covtype.binary.scale", 54, 581012);
	run_test_stat(covertype, 54, 58101, .95, 32);
	covertype = load_libsvm("./data/covtype/covtype.binary.scale", 54, 581012);
	run_test_stat(covertype, 54, 58101, .95, 64);
	covertype = load_libsvm("./data/covtype/covtype.binary.scale", 54, 581012);
	run_test_stat(covertype, 54, 58101, .95, 128);
	covertype = load_libsvm("./data/covtype/covtype.binary.scale", 54, 581012);
	run_test_stat(covertype, 54, 58101, .95, 256);


//==============================================================
/*
	// run_test_real("100leavesMar.txt", 64, 1600, .9, 50);
	//run_test_real("100leavesMar.txt", 64, 1600, .9, 192);
	//run_test_real("100leavesMar.txt", 64, 1600, .9, 16);
	//run_test_real("100leavesMar.txt", 64, 1600, .9, 8);
	//run_test_real("100leavesMar.txt", 64, 1600, .9, 4);
	run_test_real("road.txt", 1024, 10100, .9, 32);
	run_test_real("road.txt", 1024, 10100, .9, 64);
	run_test_real("road.txt", 1024, 10100, .9, 128);
	run_test_real("road.txt", 1024, 10100, .9, 256);
	//run_test_real("100leavesMar.txt", 64, 1600, .9, 193);
	run_test(3000, 256, 768);
	return 0;

	run_test_real("road.txt", 1024, 10100, .9, 8);
	run_test_real("mnist.txt", 64, 1798, .9, 8);
	run_test_real("mnist.txt", 64, 1798, .9, 64 * 3);
	run_test_real("100leavesMar.txt", 64, 1600, .9, 8);
	run_test_real("100leavesSha.txt", 64, 1600, .9, 8);
	run_test_real("100leavesTex.txt", 64, 1600, .9, 8);
	//run_test_real("road.txt", 1024, 10100, .99, 16);
	//run_test_real("road.txt", 1024, 10100, .9, 16);
	//run_test_real("road.txt", 1024, 10100, .99, 32);
	//run_test_real("road.txt", 1024, 10100, .99, 48);

	run_test(2000, 2, 2 * 3);
	run_test(5000, 2, 2 * 3);
	run_test(3000, 16, 16 * 3);
	run_test(3000, 64, 64 * 3);
	run_test(3000, 256, 256 * 3);

	run_test(2000, 2, 8);
	run_test(5000, 2, 8);
	run_test(3000, 16, 8);
	run_test(3000, 64, 8);
	run_test(3000, 256, 8);

	run_test(100000, 2, 8);
	run_test(100000, 4, 8);
	run_test(100000, 8, 8);
	run_test(100000, 10, 8);
	run_test(100000, 16, 8);

	run_test_real("road.txt", 1024, 10100, .99, 8);
	run_test_real("road.txt", 1024, 10100, .9, 8);
*/
}