about summary refs log tree commit diff
path: root/absl/random/log_uniform_int_distribution_test.cc
diff options
context:
space:
mode:
authorAbseil Team <absl-team@google.com>2019-06-21T20·11-0700
committerGennadiy Rozental <rogeeff@google.com>2019-06-21T20·18-0400
commite9324d926a9189e222741fce6e676f0944661a72 (patch)
treea08568a709940c376454da34c9d8aac021378e5f /absl/random/log_uniform_int_distribution_test.cc
parent43ef2148c0936ebf7cb4be6b19927a9d9d145b8f (diff)
Export of internal Abseil changes.
--
7a6ff16a85beb730c172d5d25cf1b5e1be885c56 by Laramie Leavitt <lar@google.com>:

Internal change.

PiperOrigin-RevId: 254454546

--
ff8f9bafaefc26d451f576ea4a06d150aed63f6f by Andy Soffer <asoffer@google.com>:

Internal changes

PiperOrigin-RevId: 254451562

--
deefc5b651b479ce36f0b4ef203e119c0c8936f2 by CJ Johnson <johnsoncj@google.com>:

Account for subtracting unsigned values from the size of InlinedVector

PiperOrigin-RevId: 254450625

--
3c677316a27bcadc17e41957c809ca472d5fef14 by Andy Soffer <asoffer@google.com>:

Add C++17's std::make_from_tuple to absl/utility/utility.h

PiperOrigin-RevId: 254411573

--
4ee3536a918830eeec402a28fc31a62c7c90b940 by CJ Johnson <johnsoncj@google.com>:

Adds benchmark for the rest of the InlinedVector public API

PiperOrigin-RevId: 254408378

--
e5a21a00700ee83498ff1efbf649169756463ee4 by CJ Johnson <johnsoncj@google.com>:

Updates the definition of InlinedVector::shrink_to_fit() to be exception safe and adds exception safety tests for it.

PiperOrigin-RevId: 254401387

--
2ea82e72b86d82d78b4e4712a63a55981b53c64b by Laramie Leavitt <lar@google.com>:

Use absl::InsecureBitGen in place of std::mt19937
in tests absl/random/...distribution_test.cc

PiperOrigin-RevId: 254289444

--
fa099e02c413a7ffda732415e8105cad26a90337 by Andy Soffer <asoffer@google.com>:

Internal changes

PiperOrigin-RevId: 254286334

--
ce34b7f36933b30cfa35b9c9a5697a792b5666e4 by Andy Soffer <asoffer@google.com>:

Internal changes

PiperOrigin-RevId: 254273059

--
6f9c473da7c2090c2e85a37c5f00622e8a912a89 by Jorg Brown <jorg@google.com>:

Change absl::container_internal::CompressedTuple to instantiate its
internal Storage class with the name of the type it's holding, rather
than the name of the Tuple.  This is not an externally-visible change,
other than less compiler memory is used and less debug information is
generated.

PiperOrigin-RevId: 254269285

--
8bd3c186bf2fc0c55d8a2dd6f28a5327502c9fba by Andy Soffer <asoffer@google.com>:

Adding short-hand IntervalClosed for IntervalClosedClosed and IntervalOpen for
IntervalOpenOpen.

PiperOrigin-RevId: 254252419

--
ea957f99b6a04fccd42aa05605605f3b44b1ecfd by Abseil Team <absl-team@google.com>:

Do not directly use __SIZEOF_INT128__.

In order to avoid linker errors when building with clang-cl (__fixunsdfti, __udivti3 and __fixunssfti are undefined), this CL uses ABSL_HAVE_INTRINSIC_INT128 which is not defined for clang-cl.

PiperOrigin-RevId: 254250739

--
89ab385cd26b34d64130bce856253aaba96d2345 by Andy Soffer <asoffer@google.com>:

Internal changes

PiperOrigin-RevId: 254242321

--
cffc793d93eca6d6bdf7de733847b6ab4a255ae9 by CJ Johnson <johnsoncj@google.com>:

Adds benchmark for InlinedVector::reserve(size_type)

PiperOrigin-RevId: 254199226

--
c90c7a9fa3c8f0c9d5114036979548b055ea2f2a by Gennadiy Rozental <rogeeff@google.com>:

Import of CCTZ from GitHub.

PiperOrigin-RevId: 254072387

--
c4c388beae016c9570ab54ffa1d52660e4a85b7b by Laramie Leavitt <lar@google.com>:

Internal cleanup.

PiperOrigin-RevId: 254062381

--
d3c992e221cc74e5372d0c8fa410170b6a43c062 by Tom Manshreck <shreck@google.com>:

Update distributions.h to Abseil standards

PiperOrigin-RevId: 254054946

--
d15ad0035c34ef11b14fadc5a4a2d3ec415f5518 by CJ Johnson <johnsoncj@google.com>:

Removes functions with only one caller from the implementation details of InlinedVector by manually inlining the definitions

PiperOrigin-RevId: 254005427

--
2f37e807efc3a8ef1f4b539bdd379917d4151520 by Andy Soffer <asoffer@google.com>:

Initial release of Abseil Random

PiperOrigin-RevId: 253999861

--
24ed1694b6430791d781ed533a8f8ccf6cac5856 by CJ Johnson <johnsoncj@google.com>:

Updates the definition of InlinedVector::assign(...)/InlinedVector::operator=(...) to new, exception-safe implementations with exception safety tests to boot

PiperOrigin-RevId: 253993691

--
5613d95f5a7e34a535cfaeadce801441e990843e by CJ Johnson <johnsoncj@google.com>:

Adds benchmarks for InlinedVector::shrink_to_fit()

PiperOrigin-RevId: 253989647

--
2a96ddfdac40bbb8cb6a7f1aeab90917067c6e63 by Abseil Team <absl-team@google.com>:

Initial release of Abseil Random

PiperOrigin-RevId: 253927497

--
bf1aff8fc9ffa921ad74643e9525ecf25b0d8dc1 by Andy Soffer <asoffer@google.com>:

Initial release of Abseil Random

PiperOrigin-RevId: 253920512

--
bfc03f4a3dcda3cf3a4b84bdb84cda24e3394f41 by Laramie Leavitt <lar@google.com>:

Internal change.

PiperOrigin-RevId: 253886486

--
05036cfcc078ca7c5f581a00dfb0daed568cbb69 by Eric Fiselier <ericwf@google.com>:

Don't include `winsock2.h` because it drags in `windows.h` and friends,
and they define awful macros like OPAQUE, ERROR, and more. This has the
potential to break abseil users.

Instead we only forward declare `timeval` and require Windows users
include `winsock2.h` themselves. This is both inconsistent and poor QoI, but so
including 'windows.h' is bad too.

PiperOrigin-RevId: 253852615
GitOrigin-RevId: 7a6ff16a85beb730c172d5d25cf1b5e1be885c56
Change-Id: Icd6aff87da26f29ec8915da856f051129987cef6
Diffstat (limited to 'absl/random/log_uniform_int_distribution_test.cc')
-rw-r--r--absl/random/log_uniform_int_distribution_test.cc277
1 files changed, 277 insertions, 0 deletions
diff --git a/absl/random/log_uniform_int_distribution_test.cc b/absl/random/log_uniform_int_distribution_test.cc
new file mode 100644
index 000000000000..0ff4c32d95d6
--- /dev/null
+++ b/absl/random/log_uniform_int_distribution_test.cc
@@ -0,0 +1,277 @@
+// 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/log_uniform_int_distribution.h"
+
+#include <cstddef>
+#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/sequence_urbg.h"
+#include "absl/random/random.h"
+#include "absl/strings/str_cat.h"
+#include "absl/strings/str_format.h"
+#include "absl/strings/str_replace.h"
+#include "absl/strings/strip.h"
+
+namespace {
+
+template <typename IntType>
+class LogUniformIntDistributionTypeTest : public ::testing::Test {};
+
+using IntTypes = ::testing::Types<int8_t, int16_t, int32_t, int64_t,  //
+                                  uint8_t, uint16_t, uint32_t, uint64_t>;
+TYPED_TEST_CASE(LogUniformIntDistributionTypeTest, IntTypes);
+
+TYPED_TEST(LogUniformIntDistributionTypeTest, SerializeTest) {
+  using param_type =
+      typename absl::log_uniform_int_distribution<TypeParam>::param_type;
+  using Limits = std::numeric_limits<TypeParam>;
+
+  constexpr int kCount = 1000;
+  absl::InsecureBitGen gen;
+  for (const auto& param : {
+           param_type(0, 1),                             //
+           param_type(0, 2),                             //
+           param_type(0, 2, 10),                         //
+           param_type(9, 32, 4),                         //
+           param_type(1, 101, 10),                       //
+           param_type(1, Limits::max() / 2),             //
+           param_type(0, Limits::max() - 1),             //
+           param_type(0, Limits::max(), 2),              //
+           param_type(0, Limits::max(), 10),             //
+           param_type(Limits::min(), 0),                 //
+           param_type(Limits::lowest(), Limits::max()),  //
+           param_type(Limits::min(), Limits::max()),     //
+       }) {
+    // Validate parameters.
+    const auto min = param.min();
+    const auto max = param.max();
+    const auto base = param.base();
+    absl::log_uniform_int_distribution<TypeParam> before(min, max, base);
+    EXPECT_EQ(before.min(), param.min());
+    EXPECT_EQ(before.max(), param.max());
+    EXPECT_EQ(before.base(), param.base());
+
+    {
+      absl::log_uniform_int_distribution<TypeParam> via_param(param);
+      EXPECT_EQ(via_param, before);
+    }
+
+    // Validate stream serialization.
+    std::stringstream ss;
+    ss << before;
+
+    absl::log_uniform_int_distribution<TypeParam> after(3, 6, 17);
+
+    EXPECT_NE(before.max(), after.max());
+    EXPECT_NE(before.base(), after.base());
+    EXPECT_NE(before.param(), after.param());
+    EXPECT_NE(before, after);
+
+    ss >> after;
+
+    EXPECT_EQ(before.min(), after.min());
+    EXPECT_EQ(before.max(), after.max());
+    EXPECT_EQ(before.base(), after.base());
+    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);
+      EXPECT_GE(sample, after.min());
+      EXPECT_LE(sample, after.max());
+      if (sample > sample_max) sample_max = sample;
+      if (sample < sample_min) sample_min = sample;
+    }
+    ABSL_INTERNAL_LOG(INFO,
+                      absl::StrCat("Range: ", +sample_min, ", ", +sample_max));
+  }
+}
+
+using log_uniform_i32 = absl::log_uniform_int_distribution<int32_t>;
+
+class LogUniformIntChiSquaredTest
+    : public testing::TestWithParam<log_uniform_i32::param_type> {
+ public:
+  // The ChiSquaredTestImpl provides a chi-squared goodness of fit test for
+  // data generated by the log-uniform-int distribution.
+  double ChiSquaredTestImpl();
+
+  absl::InsecureBitGen rng_;
+};
+
+double LogUniformIntChiSquaredTest::ChiSquaredTestImpl() {
+  using absl::random_internal::kChiSquared;
+
+  const auto& param = GetParam();
+
+  // Check the distribution of L=log(log_uniform_int_distribution, base),
+  // expecting that L is roughly uniformly distributed, that is:
+  //
+  //   P[L=0] ~= P[L=1] ~= ... ~= P[L=log(max)]
+  //
+  // For a total of X entries, each bucket should contain some number of samples
+  // in the interval [X/k - a, X/k + a].
+  //
+  // Where `a` is approximately sqrt(X/k). This is validated by bucketing
+  // according to the log function and using a chi-squared test for uniformity.
+
+  const bool is_2 = (param.base() == 2);
+  const double base_log = 1.0 / std::log(param.base());
+  const auto bucket_index = [base_log, is_2, &param](int32_t x) {
+    uint64_t y = static_cast<uint64_t>(x) - param.min();
+    return (y == 0) ? 0
+                    : is_2 ? static_cast<int>(1 + std::log2(y))
+                           : static_cast<int>(1 + std::log(y) * base_log);
+  };
+  const int max_bucket = bucket_index(param.max());  // inclusive
+  const size_t trials = 15 + (max_bucket + 1) * 10;
+
+  log_uniform_i32 dist(param);
+
+  std::vector<int64_t> buckets(max_bucket + 1);
+  for (size_t i = 0; i < trials; ++i) {
+    const auto sample = dist(rng_);
+    // Check the bounds.
+    ABSL_ASSERT(sample <= dist.max());
+    ABSL_ASSERT(sample >= dist.min());
+    // Convert the output of the generator to one of num_bucket buckets.
+    int bucket = bucket_index(sample);
+    ABSL_ASSERT(bucket <= max_bucket);
+    ++buckets[bucket];
+  }
+
+  // The null-hypothesis is that the distribution is uniform with respect to
+  // log-uniform-int bucketization.
+  const int dof = buckets.size() - 1;
+  const double expected = trials / static_cast<double>(buckets.size());
+
+  const double threshold = absl::random_internal::ChiSquareValue(dof, 0.98);
+
+  double chi_square = absl::random_internal::ChiSquareWithExpected(
+      std::begin(buckets), std::end(buckets), expected);
+
+  const double p = absl::random_internal::ChiSquarePValue(chi_square, dof);
+
+  if (chi_square > threshold) {
+    ABSL_INTERNAL_LOG(INFO, "values");
+    for (size_t i = 0; i < buckets.size(); i++) {
+      ABSL_INTERNAL_LOG(INFO, absl::StrCat(i, ": ", buckets[i]));
+    }
+    ABSL_INTERNAL_LOG(INFO,
+                      absl::StrFormat("trials=%d\n"
+                                      "%s(data, %d) = %f (%f)\n"
+                                      "%s @ 0.98 = %f",
+                                      trials, kChiSquared, dof, chi_square, p,
+                                      kChiSquared, threshold));
+  }
+  return p;
+}
+
+TEST_P(LogUniformIntChiSquaredTest, MultiTest) {
+  const int kTrials = 5;
+
+  int failures = 0;
+  for (int i = 0; i < kTrials; i++) {
+    double p_value = ChiSquaredTestImpl();
+    if (p_value < 0.005) {
+      failures++;
+    }
+  }
+
+  // There is a 0.10% chance of producing at least one failure, so raise the
+  // failure threshold high enough to allow for a flake rate < 10,000.
+  EXPECT_LE(failures, 4);
+}
+
+// Generate the parameters for the test.
+std::vector<log_uniform_i32::param_type> GenParams() {
+  using Param = log_uniform_i32::param_type;
+  using Limits = std::numeric_limits<int32_t>;
+
+  return std::vector<Param>{
+      Param{0, 1, 2},
+      Param{1, 1, 2},
+      Param{0, 2, 2},
+      Param{0, 3, 2},
+      Param{0, 4, 2},
+      Param{0, 9, 10},
+      Param{0, 10, 10},
+      Param{0, 11, 10},
+      Param{1, 10, 10},
+      Param{0, (1 << 8) - 1, 2},
+      Param{0, (1 << 8), 2},
+      Param{0, (1 << 30) - 1, 2},
+      Param{-1000, 1000, 10},
+      Param{0, Limits::max(), 2},
+      Param{0, Limits::max(), 3},
+      Param{0, Limits::max(), 10},
+      Param{Limits::min(), 0},
+      Param{Limits::min(), Limits::max(), 2},
+  };
+}
+
+std::string ParamName(
+    const ::testing::TestParamInfo<log_uniform_i32::param_type>& info) {
+  const auto& p = info.param;
+  std::string name =
+      absl::StrCat("min_", p.min(), "__max_", p.max(), "__base_", p.base());
+  return absl::StrReplaceAll(name, {{"+", "_"}, {"-", "_"}, {".", "_"}});
+}
+
+INSTANTIATE_TEST_SUITE_P(, LogUniformIntChiSquaredTest,
+                         ::testing::ValuesIn(GenParams()), ParamName);
+
+// NOTE: absl::log_uniform_int_distribution is not guaranteed to be stable.
+TEST(LogUniformIntDistributionTest, StabilityTest) {
+  using testing::ElementsAre;
+  // absl::uniform_int_distribution stability relies on
+  // absl::random_internal::LeadingSetBit, std::log, std::pow.
+  absl::random_internal::sequence_urbg urbg(
+      {0x0003eb76f6f7f755ull, 0xFFCEA50FDB2F953Bull, 0xC332DDEFBE6C5AA5ull,
+       0x6558218568AB9702ull, 0x2AEF7DAD5B6E2F84ull, 0x1521B62829076170ull,
+       0xECDD4775619F1510ull, 0x13CCA830EB61BD96ull, 0x0334FE1EAA0363CFull,
+       0xB5735C904C70A239ull, 0xD59E9E0BCBAADE14ull, 0xEECC86BC60622CA7ull});
+
+  std::vector<int> output(6);
+
+  {
+    absl::log_uniform_int_distribution<int32_t> dist(0, 256);
+    std::generate(std::begin(output), std::end(output),
+                  [&] { return dist(urbg); });
+    EXPECT_THAT(output, ElementsAre(256, 66, 4, 6, 57, 103));
+  }
+  urbg.reset();
+  {
+    absl::log_uniform_int_distribution<int32_t> dist(0, 256, 10);
+    std::generate(std::begin(output), std::end(output),
+                  [&] { return dist(urbg); });
+    EXPECT_THAT(output, ElementsAre(8, 4, 0, 0, 0, 69));
+  }
+}
+
+}  // namespace