summaryrefslogtreecommitdiff
path: root/absl/random/uniform_real_distribution_test.cc
blob: 8cf874d61c5cd34068a84057e805227ebc9af709 (plain)
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
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
// Copyright 2017 The Abseil Authors.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
//      https://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.

#include "absl/random/uniform_real_distribution.h"

#include <cmath>
#include <cstdint>
#include <iterator>
#include <random>
#include <sstream>
#include <string>
#include <vector>

#include "gmock/gmock.h"
#include "gtest/gtest.h"
#include "absl/base/internal/raw_logging.h"
#include "absl/random/internal/chi_square.h"
#include "absl/random/internal/distribution_test_util.h"
#include "absl/random/internal/pcg_engine.h"
#include "absl/random/internal/sequence_urbg.h"
#include "absl/random/random.h"
#include "absl/strings/str_cat.h"

// NOTES:
// * Some documentation on generating random real values suggests that
//   it is possible to use std::nextafter(b, DBL_MAX) to generate a value on
//   the closed range [a, b]. Unfortunately, that technique is not universally
//   reliable due to floating point quantization.
//
// * absl::uniform_real_distribution<float> generates between 2^28 and 2^29
//   distinct floating point values in the range [0, 1).
//
// * absl::uniform_real_distribution<float> generates at least 2^23 distinct
//   floating point values in the range [1, 2). This should be the same as
//   any other range covered by a single exponent in IEEE 754.
//
// * absl::uniform_real_distribution<double> generates more than 2^52 distinct
//   values in the range [0, 1), and should generate at least 2^52 distinct
//   values in the range of [1, 2).
//

namespace {

template <typename RealType>
class UniformRealDistributionTest : public ::testing::Test {};

using RealTypes = ::testing::Types<float, double, long double>;

TYPED_TEST_SUITE(UniformRealDistributionTest, RealTypes);

TYPED_TEST(UniformRealDistributionTest, ParamSerializeTest) {
  using param_type =
      typename absl::uniform_real_distribution<TypeParam>::param_type;

  constexpr const TypeParam a{1152921504606846976};

  constexpr int kCount = 1000;
  absl::InsecureBitGen gen;
  for (const auto& param : {
           param_type(),
           param_type(TypeParam(2.0), TypeParam(2.0)),  // Same
           param_type(TypeParam(-0.1), TypeParam(0.1)),
           param_type(TypeParam(0.05), TypeParam(0.12)),
           param_type(TypeParam(-0.05), TypeParam(0.13)),
           param_type(TypeParam(-0.05), TypeParam(-0.02)),
           // double range = 0
           // 2^60 , 2^60 + 2^6
           param_type(a, TypeParam(1152921504606847040)),
           // 2^60 , 2^60 + 2^7
           param_type(a, TypeParam(1152921504606847104)),
           // double range = 2^8
           // 2^60 , 2^60 + 2^8
           param_type(a, TypeParam(1152921504606847232)),
           // float range = 0
           // 2^60 , 2^60 + 2^36
           param_type(a, TypeParam(1152921573326323712)),
           // 2^60 , 2^60 + 2^37
           param_type(a, TypeParam(1152921642045800448)),
           // float range = 2^38
           // 2^60 , 2^60 + 2^38
           param_type(a, TypeParam(1152921779484753920)),
           // Limits
           param_type(0, std::numeric_limits<TypeParam>::max()),
           param_type(std::numeric_limits<TypeParam>::lowest(), 0),
           param_type(0, std::numeric_limits<TypeParam>::epsilon()),
           param_type(-std::numeric_limits<TypeParam>::epsilon(),
                      std::numeric_limits<TypeParam>::epsilon()),
           param_type(std::numeric_limits<TypeParam>::epsilon(),
                      2 * std::numeric_limits<TypeParam>::epsilon()),
       }) {
    // Validate parameters.
    const auto a = param.a();
    const auto b = param.b();
    absl::uniform_real_distribution<TypeParam> before(a, b);
    EXPECT_EQ(before.a(), param.a());
    EXPECT_EQ(before.b(), param.b());

    {
      absl::uniform_real_distribution<TypeParam> via_param(param);
      EXPECT_EQ(via_param, before);
    }

    std::stringstream ss;
    ss << before;
    absl::uniform_real_distribution<TypeParam> after(TypeParam(1.0),
                                                     TypeParam(3.1));

    EXPECT_NE(before.a(), after.a());
    EXPECT_NE(before.b(), after.b());
    EXPECT_NE(before.param(), after.param());
    EXPECT_NE(before, after);

    ss >> after;

    EXPECT_EQ(before.a(), after.a());
    EXPECT_EQ(before.b(), after.b());
    EXPECT_EQ(before.param(), after.param());
    EXPECT_EQ(before, after);

    // Smoke test.
    auto sample_min = after.max();
    auto sample_max = after.min();
    for (int i = 0; i < kCount; i++) {
      auto sample = after(gen);
      // Failure here indicates a bug in uniform_real_distribution::operator(),
      // or bad parameters--range too large, etc.
      if (after.min() == after.max()) {
        EXPECT_EQ(sample, after.min());
      } else {
        EXPECT_GE(sample, after.min());
        EXPECT_LT(sample, after.max());
      }
      if (sample > sample_max) {
        sample_max = sample;
      }
      if (sample < sample_min) {
        sample_min = sample;
      }
    }

    if (!std::is_same<TypeParam, long double>::value) {
      // static_cast<double>(long double) can overflow.
      std::string msg = absl::StrCat("Range: ", static_cast<double>(sample_min),
                                     ", ", static_cast<double>(sample_max));
      ABSL_RAW_LOG(INFO, "%s", msg.c_str());
    }
  }
}

#ifdef _MSC_VER
#pragma warning(push)
#pragma warning(disable:4756)  // Constant arithmetic overflow.
#endif
TYPED_TEST(UniformRealDistributionTest, ViolatesPreconditionsDeathTest) {
#if GTEST_HAS_DEATH_TEST
  // Hi < Lo
  EXPECT_DEBUG_DEATH(
      { absl::uniform_real_distribution<TypeParam> dist(10.0, 1.0); }, "");

  // Hi - Lo > numeric_limits<>::max()
  EXPECT_DEBUG_DEATH(
      {
        absl::uniform_real_distribution<TypeParam> dist(
            std::numeric_limits<TypeParam>::lowest(),
            std::numeric_limits<TypeParam>::max());
      },
      "");
#endif  // GTEST_HAS_DEATH_TEST
#if defined(NDEBUG)
  // opt-mode, for invalid parameters, will generate a garbage value,
  // but should not enter an infinite loop.
  absl::InsecureBitGen gen;
  {
    absl::uniform_real_distribution<TypeParam> dist(10.0, 1.0);
    auto x = dist(gen);
    EXPECT_FALSE(std::isnan(x)) << x;
  }
  {
    absl::uniform_real_distribution<TypeParam> dist(
        std::numeric_limits<TypeParam>::lowest(),
        std::numeric_limits<TypeParam>::max());
    auto x = dist(gen);
    // Infinite result.
    EXPECT_FALSE(std::isfinite(x)) << x;
  }
#endif  // NDEBUG
}
#ifdef _MSC_VER
#pragma warning(pop)  // warning(disable:4756)
#endif

TYPED_TEST(UniformRealDistributionTest, TestMoments) {
  constexpr int kSize = 1000000;
  std::vector<double> values(kSize);

  // We use a fixed bit generator for distribution accuracy tests.  This allows
  // these tests to be deterministic, while still testing the qualify of the
  // implementation.
  absl::random_internal::pcg64_2018_engine rng{0x2B7E151628AED2A6};

  absl::uniform_real_distribution<TypeParam> dist;
  for (int i = 0; i < kSize; i++) {
    values[i] = dist(rng);
  }

  const auto moments =
      absl::random_internal::ComputeDistributionMoments(values);
  EXPECT_NEAR(0.5, moments.mean, 0.01);
  EXPECT_NEAR(1 / 12.0, moments.variance, 0.015);
  EXPECT_NEAR(0.0, moments.skewness, 0.02);
  EXPECT_NEAR(9 / 5.0, moments.kurtosis, 0.015);
}

TYPED_TEST(UniformRealDistributionTest, ChiSquaredTest50) {
  using absl::random_internal::kChiSquared;
  using param_type =
      typename absl::uniform_real_distribution<TypeParam>::param_type;

  constexpr size_t kTrials = 100000;
  constexpr int kBuckets = 50;
  constexpr double kExpected =
      static_cast<double>(kTrials) / static_cast<double>(kBuckets);

  // 1-in-100000 threshold, but remember, there are about 8 tests
  // in this file. And the test could fail for other reasons.
  // Empirically validated with --runs_per_test=10000.
  const int kThreshold =
      absl::random_internal::ChiSquareValue(kBuckets - 1, 0.999999);

  // We use a fixed bit generator for distribution accuracy tests.  This allows
  // these tests to be deterministic, while still testing the qualify of the
  // implementation.
  absl::random_internal::pcg64_2018_engine rng{0x2B7E151628AED2A6};

  for (const auto& param : {param_type(0, 1), param_type(5, 12),
                            param_type(-5, 13), param_type(-5, -2)}) {
    const double min_val = param.a();
    const double max_val = param.b();
    const double factor = kBuckets / (max_val - min_val);

    std::vector<int32_t> counts(kBuckets, 0);
    absl::uniform_real_distribution<TypeParam> dist(param);
    for (size_t i = 0; i < kTrials; i++) {
      auto x = dist(rng);
      auto bucket = static_cast<size_t>((x - min_val) * factor);
      counts[bucket]++;
    }

    double chi_square = absl::random_internal::ChiSquareWithExpected(
        std::begin(counts), std::end(counts), kExpected);
    if (chi_square > kThreshold) {
      double p_value =
          absl::random_internal::ChiSquarePValue(chi_square, kBuckets);

      // Chi-squared test failed. Output does not appear to be uniform.
      std::string msg;
      for (const auto& a : counts) {
        absl::StrAppend(&msg, a, "\n");
      }
      absl::StrAppend(&msg, kChiSquared, " p-value ", p_value, "\n");
      absl::StrAppend(&msg, "High ", kChiSquared, " value: ", chi_square, " > ",
                      kThreshold);
      ABSL_RAW_LOG(INFO, "%s", msg.c_str());
      FAIL() << msg;
    }
  }
}

TYPED_TEST(UniformRealDistributionTest, StabilityTest) {
  // absl::uniform_real_distribution stability relies only on
  // random_internal::RandU64ToDouble and random_internal::RandU64ToFloat.
  absl::random_internal::sequence_urbg urbg(
      {0x0003eb76f6f7f755ull, 0xFFCEA50FDB2F953Bull, 0xC332DDEFBE6C5AA5ull,
       0x6558218568AB9702ull, 0x2AEF7DAD5B6E2F84ull, 0x1521B62829076170ull,
       0xECDD4775619F1510ull, 0x13CCA830EB61BD96ull, 0x0334FE1EAA0363CFull,
       0xB5735C904C70A239ull, 0xD59E9E0BCBAADE14ull, 0xEECC86BC60622CA7ull});

  std::vector<int> output(12);

  absl::uniform_real_distribution<TypeParam> dist;
  std::generate(std::begin(output), std::end(output), [&] {
    return static_cast<int>(TypeParam(1000000) * dist(urbg));
  });

  EXPECT_THAT(
      output,  //
      testing::ElementsAre(59, 999246, 762494, 395876, 167716, 82545, 925251,
                           77341, 12527, 708791, 834451, 932808));
}

TEST(UniformRealDistributionTest, AlgorithmBounds) {
  absl::uniform_real_distribution<double> dist;

  {
    // This returns the smallest value >0 from absl::uniform_real_distribution.
    absl::random_internal::sequence_urbg urbg({0x0000000000000001ull});
    double a = dist(urbg);
    EXPECT_EQ(a, 5.42101086242752217004e-20);
  }

  {
    // This returns a value very near 0.5 from absl::uniform_real_distribution.
    absl::random_internal::sequence_urbg urbg({0x7fffffffffffffefull});
    double a = dist(urbg);
    EXPECT_EQ(a, 0.499999999999999944489);
  }
  {
    // This returns a value very near 0.5 from absl::uniform_real_distribution.
    absl::random_internal::sequence_urbg urbg({0x8000000000000000ull});
    double a = dist(urbg);
    EXPECT_EQ(a, 0.5);
  }

  {
    // This returns the largest value <1 from absl::uniform_real_distribution.
    absl::random_internal::sequence_urbg urbg({0xFFFFFFFFFFFFFFEFull});
    double a = dist(urbg);
    EXPECT_EQ(a, 0.999999999999999888978);
  }
  {
    // This *ALSO* returns the largest value <1.
    absl::random_internal::sequence_urbg urbg({0xFFFFFFFFFFFFFFFFull});
    double a = dist(urbg);
    EXPECT_EQ(a, 0.999999999999999888978);
  }
}

}  // namespace