• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  *  Copyright 2011 The WebRTC Project Authors. All rights reserved.
3  *
4  *  Use of this source code is governed by a BSD-style license
5  *  that can be found in the LICENSE file in the root of the source
6  *  tree. An additional intellectual property rights grant can be found
7  *  in the file PATENTS.  All contributing project authors may
8  *  be found in the AUTHORS file in the root of the source tree.
9  */
10 
11 #include "rtc_base/rolling_accumulator.h"
12 
13 #include <random>
14 
15 #include "test/gtest.h"
16 
17 namespace rtc {
18 
19 namespace {
20 
21 const double kLearningRate = 0.5;
22 
23 // Add |n| samples drawn from uniform distribution in [a;b].
FillStatsFromUniformDistribution(RollingAccumulator<double> & stats,int n,double a,double b)24 void FillStatsFromUniformDistribution(RollingAccumulator<double>& stats,
25                                       int n,
26                                       double a,
27                                       double b) {
28   std::mt19937 gen{std::random_device()()};
29   std::uniform_real_distribution<> dis(a, b);
30 
31   for (int i = 1; i <= n; i++) {
32     stats.AddSample(dis(gen));
33   }
34 }
35 }  // namespace
36 
TEST(RollingAccumulatorTest,ZeroSamples)37 TEST(RollingAccumulatorTest, ZeroSamples) {
38   RollingAccumulator<int> accum(10);
39 
40   EXPECT_EQ(0U, accum.count());
41   EXPECT_DOUBLE_EQ(0.0, accum.ComputeMean());
42   EXPECT_DOUBLE_EQ(0.0, accum.ComputeVariance());
43   EXPECT_EQ(0, accum.ComputeMin());
44   EXPECT_EQ(0, accum.ComputeMax());
45 }
46 
TEST(RollingAccumulatorTest,SomeSamples)47 TEST(RollingAccumulatorTest, SomeSamples) {
48   RollingAccumulator<int> accum(10);
49   for (int i = 0; i < 4; ++i) {
50     accum.AddSample(i);
51   }
52 
53   EXPECT_EQ(4U, accum.count());
54   EXPECT_DOUBLE_EQ(1.5, accum.ComputeMean());
55   EXPECT_NEAR(2.26666, accum.ComputeWeightedMean(kLearningRate), 0.01);
56   EXPECT_DOUBLE_EQ(1.25, accum.ComputeVariance());
57   EXPECT_EQ(0, accum.ComputeMin());
58   EXPECT_EQ(3, accum.ComputeMax());
59 }
60 
TEST(RollingAccumulatorTest,RollingSamples)61 TEST(RollingAccumulatorTest, RollingSamples) {
62   RollingAccumulator<int> accum(10);
63   for (int i = 0; i < 12; ++i) {
64     accum.AddSample(i);
65   }
66 
67   EXPECT_EQ(10U, accum.count());
68   EXPECT_DOUBLE_EQ(6.5, accum.ComputeMean());
69   EXPECT_NEAR(10.0, accum.ComputeWeightedMean(kLearningRate), 0.01);
70   EXPECT_NEAR(9.0, accum.ComputeVariance(), 1.0);
71   EXPECT_EQ(2, accum.ComputeMin());
72   EXPECT_EQ(11, accum.ComputeMax());
73 }
74 
TEST(RollingAccumulatorTest,ResetSamples)75 TEST(RollingAccumulatorTest, ResetSamples) {
76   RollingAccumulator<int> accum(10);
77 
78   for (int i = 0; i < 10; ++i) {
79     accum.AddSample(100);
80   }
81   EXPECT_EQ(10U, accum.count());
82   EXPECT_DOUBLE_EQ(100.0, accum.ComputeMean());
83   EXPECT_EQ(100, accum.ComputeMin());
84   EXPECT_EQ(100, accum.ComputeMax());
85 
86   accum.Reset();
87   EXPECT_EQ(0U, accum.count());
88 
89   for (int i = 0; i < 5; ++i) {
90     accum.AddSample(i);
91   }
92 
93   EXPECT_EQ(5U, accum.count());
94   EXPECT_DOUBLE_EQ(2.0, accum.ComputeMean());
95   EXPECT_EQ(0, accum.ComputeMin());
96   EXPECT_EQ(4, accum.ComputeMax());
97 }
98 
TEST(RollingAccumulatorTest,RollingSamplesDouble)99 TEST(RollingAccumulatorTest, RollingSamplesDouble) {
100   RollingAccumulator<double> accum(10);
101   for (int i = 0; i < 23; ++i) {
102     accum.AddSample(5 * i);
103   }
104 
105   EXPECT_EQ(10u, accum.count());
106   EXPECT_DOUBLE_EQ(87.5, accum.ComputeMean());
107   EXPECT_NEAR(105.049, accum.ComputeWeightedMean(kLearningRate), 0.1);
108   EXPECT_NEAR(229.166667, accum.ComputeVariance(), 25);
109   EXPECT_DOUBLE_EQ(65.0, accum.ComputeMin());
110   EXPECT_DOUBLE_EQ(110.0, accum.ComputeMax());
111 }
112 
TEST(RollingAccumulatorTest,ComputeWeightedMeanCornerCases)113 TEST(RollingAccumulatorTest, ComputeWeightedMeanCornerCases) {
114   RollingAccumulator<int> accum(10);
115   EXPECT_DOUBLE_EQ(0.0, accum.ComputeWeightedMean(kLearningRate));
116   EXPECT_DOUBLE_EQ(0.0, accum.ComputeWeightedMean(0.0));
117   EXPECT_DOUBLE_EQ(0.0, accum.ComputeWeightedMean(1.1));
118 
119   for (int i = 0; i < 8; ++i) {
120     accum.AddSample(i);
121   }
122 
123   EXPECT_DOUBLE_EQ(3.5, accum.ComputeMean());
124   EXPECT_DOUBLE_EQ(3.5, accum.ComputeWeightedMean(0));
125   EXPECT_DOUBLE_EQ(3.5, accum.ComputeWeightedMean(1.1));
126   EXPECT_NEAR(6.0, accum.ComputeWeightedMean(kLearningRate), 0.1);
127 }
128 
TEST(RollingAccumulatorTest,VarianceFromUniformDistribution)129 TEST(RollingAccumulatorTest, VarianceFromUniformDistribution) {
130   // Check variance converge to 1/12 for [0;1) uniform distribution.
131   // Acts as a sanity check for NumericStabilityForVariance test.
132   RollingAccumulator<double> stats(/*max_count=*/0.5e6);
133   FillStatsFromUniformDistribution(stats, 1e6, 0, 1);
134 
135   EXPECT_NEAR(stats.ComputeVariance(), 1. / 12, 1e-3);
136 }
137 
TEST(RollingAccumulatorTest,NumericStabilityForVariance)138 TEST(RollingAccumulatorTest, NumericStabilityForVariance) {
139   // Same test as VarianceFromUniformDistribution,
140   // except the range is shifted to [1e9;1e9+1).
141   // Variance should also converge to 1/12.
142   // NB: Although we lose precision for the samples themselves, the fractional
143   //     part still enjoys 22 bits of mantissa and errors should even out,
144   //     so that couldn't explain a mismatch.
145   RollingAccumulator<double> stats(/*max_count=*/0.5e6);
146   FillStatsFromUniformDistribution(stats, 1e6, 1e9, 1e9 + 1);
147 
148   EXPECT_NEAR(stats.ComputeVariance(), 1. / 12, 1e-3);
149 }
150 }  // namespace rtc
151