datasketches 0.4.3 → 0.5.0

data/vendor/datasketches-cpp/filters/test/bloom_filter_test.cpp

@@ -0,0 +1,406 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements.  See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership.  The ASF licenses this file
+ * to you 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
+ *
+ *   http://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 <catch2/catch.hpp>
+
+#include "bloom_filter.hpp"
+
+#ifdef TEST_BINARY_INPUT_PATH
+static std::string testBinaryInputPath = TEST_BINARY_INPUT_PATH;
+#else
+static std::string testBinaryInputPath = "test/";
+#endif
+
+namespace datasketches {
+
+TEST_CASE("bloom_filter: invalid constructor args", "[bloom_filter]") {
+  REQUIRE_THROWS_AS(bloom_filter::builder::create_by_size(0, 4), std::invalid_argument);
+  REQUIRE_THROWS_AS(bloom_filter::builder::create_by_size(1L << 60, 4), std::invalid_argument);
+  REQUIRE_THROWS_AS(bloom_filter::builder::create_by_size(65535, 0), std::invalid_argument);
+}
+
+TEST_CASE("bloom_filter: standard constructors", "[bloom_filter]") {
+  uint64_t num_items = 4000;
+  double fpp = 0.01;
+
+  uint64_t num_bits = bloom_filter::builder::suggest_num_filter_bits(num_items, fpp);
+  uint16_t num_hashes = bloom_filter::builder::suggest_num_hashes(num_items, num_bits);
+  uint64_t seed = 89023;
+
+  auto bf = bloom_filter::builder::create_by_size(num_bits, num_hashes, seed);
+  uint64_t adjusted_num_bits = (num_bits + 63) & ~0x3F; // round up to the nearest multiple of 64
+  REQUIRE(bf.get_capacity() == adjusted_num_bits);
+  REQUIRE(bf.get_num_hashes() == num_hashes);
+  REQUIRE(bf.get_seed() == seed);
+  REQUIRE(bf.is_empty());
+
+  // should match above
+  bf = bloom_filter::builder::create_by_accuracy(num_items, fpp, seed);
+  REQUIRE(bf.get_capacity() == adjusted_num_bits);
+  REQUIRE(bf.get_num_hashes() == num_hashes);
+  REQUIRE(bf.get_seed() == seed);
+  REQUIRE(bf.is_empty());
+
+  // same for initializing memory in-place
+  size_t serialized_size_bytes = bloom_filter::get_serialized_size_bytes(num_bits);
+  uint8_t* bytes = new uint8_t[serialized_size_bytes];
+
+  bf = bloom_filter::builder::initialize_by_size(bytes, serialized_size_bytes, num_bits, num_hashes, seed);
+  REQUIRE(bf.get_capacity() == adjusted_num_bits);
+  REQUIRE(bf.get_num_hashes() == num_hashes);
+  REQUIRE(bf.get_seed() == seed);
+  REQUIRE(bf.is_empty());
+
+  bf = bloom_filter::builder::initialize_by_accuracy(bytes, serialized_size_bytes, num_items, fpp, seed);
+  REQUIRE(bf.get_capacity() == adjusted_num_bits);
+  REQUIRE(bf.get_num_hashes() == num_hashes);
+  REQUIRE(bf.get_seed() == seed);
+  REQUIRE(bf.is_empty());
+
+  delete [] bytes;
+}
+
+TEST_CASE("bloom_filter: basic operations", "[bloom_filter]") {
+  uint64_t num_items = 5000;
+  double fpp = 0.01;
+  uint64_t seed = 4897301548054ULL;
+
+  auto bf = bloom_filter::builder::create_by_accuracy(num_items, fpp, seed);
+  REQUIRE(bf.is_empty());
+  REQUIRE(bf.get_bits_used() == 0);
+
+  for (uint64_t i = 0; i < num_items; ++i) {
+    bf.query_and_update(i);
+  }
+
+  REQUIRE(!bf.is_empty());
+  // filter is about 50% full at target capacity
+  // since seed is fixed we expect an exact value every time
+  // but leaving the approximate test in since that's more the "expectation"
+  REQUIRE(bf.get_bits_used() == 24793); // exact value is not important but should be consistent
+  REQUIRE(bf.get_bits_used() == Approx(0.5 * bf.get_capacity()).epsilon(0.05)); // just over 3.3% in practice
+
+  uint32_t num_found = 0;
+  for (uint64_t i = num_items; i < bf.get_capacity(); ++i) {
+    if (bf.query(i)) {
+      ++num_found;
+    }
+  }
+  // fpp is average with significant variance -- even at 12% it would fail occasionally
+  REQUIRE(num_found == 423);
+  //REQUIRE(num_found == Approx((bf.get_capacity() - num_items) * fpp).epsilon(0.12));
+  auto bytes = bf.serialize();
+
+  // initialize in memory and run the same tests
+  // also checking against the results from the first part
+  uint8_t* bf_memory = new uint8_t[bytes.size()];
+  auto bf2 = bloom_filter::builder::initialize_by_accuracy(bf_memory, bytes.size(), num_items, fpp, bf.get_seed());
+  REQUIRE(bf2.is_empty());
+  REQUIRE(bf2.get_bits_used() == 0);
+
+  for (uint64_t i = 0; i < num_items; ++i) {
+    bf2.query_and_update(i);
+  }
+
+  REQUIRE(!bf2.is_empty());
+  REQUIRE(bf2.get_bits_used() == bf.get_bits_used()); // should exactly match above
+
+  uint32_t num_found2 = 0;
+  for (uint64_t i = num_items; i < bf2.get_capacity(); ++i) {
+    if (bf2.query(i)) {
+      ++num_found2;
+    }
+  }
+  REQUIRE(num_found == num_found2); // should exactly match above
+  auto bytes2 = bf2.serialize();
+
+  REQUIRE(bytes.size() == bytes2.size());
+  for (size_t i = 0; i < bytes.size(); ++i) {
+    REQUIRE(bytes[i] == bytes2[i]);
+  }
+
+  // check that raw memory also matches serialized sketch
+  const uint8_t* bf_bytes = bf2.get_wrapped_memory();
+  REQUIRE(bf_bytes == bf_memory);
+  for (size_t i = 0; i < bytes.size(); ++i) {
+    REQUIRE(bf_bytes[i] == bytes[i]);
+  }
+
+  // ensure the filters reset properly
+  bf.reset();
+  REQUIRE(bf.is_empty());
+  REQUIRE(bf.get_bits_used() == 0);
+
+  bf2.reset();
+  REQUIRE(bf2.is_empty());
+  REQUIRE(bf2.get_bits_used() == 0);
+
+  delete [] bf_memory;
+}
+
+TEST_CASE("bloom_filter: inversion", "[bloom_filter]") {
+  uint64_t num_bits = 8192;
+  uint16_t num_hashes = 3;
+
+  auto bf = bloom_filter::builder::create_by_size(num_bits, num_hashes);
+
+  uint64_t n = 500;
+  for (uint64_t i = 0; i < n; ++i) {
+    bf.update(i);
+  }
+  uint64_t num_bits_set = bf.get_bits_used();
+  bf.invert();
+  REQUIRE(bf.get_bits_used() == num_bits - num_bits_set);
+
+  // original items should be mostly not-present
+  uint32_t num_found = 0;
+  for (uint64_t i = 0; i < n; ++i) {
+    if (bf.query(i)) {
+      ++num_found;
+    }
+  }
+  REQUIRE(num_found < n / 10);
+
+  // many other items should be "present"
+  num_found = 0;
+  for (uint64_t i = n; i < num_bits; ++i) {
+    if (bf.query(i)) {
+      ++num_found;
+    }
+  }
+  REQUIRE(num_found > n);
+}
+
+TEST_CASE("bloom_filter: incompatible set operations", "[bloom_filter]") {
+  uint64_t num_bits = 32768;
+  uint16_t num_hashes = 4;
+
+  auto bf1 = bloom_filter::builder::create_by_size(num_bits, num_hashes);
+
+  // mismatched num bits
+  auto bf2 = bloom_filter::builder::create_by_size(2 * num_bits, num_hashes);
+  REQUIRE_THROWS_AS(bf1.union_with(bf2), std::invalid_argument);
+
+  // mismatched num hashes
+  auto bf3 = bloom_filter::builder::create_by_size(num_bits, 2 * num_hashes);
+  REQUIRE_THROWS_AS(bf1.intersect(bf2), std::invalid_argument);
+
+  // mismatched seed
+  auto bf4 = bloom_filter::builder::create_by_size(num_bits, num_hashes, bf1.get_seed() + 1);
+  REQUIRE_THROWS_AS(bf1.union_with(bf4), std::invalid_argument);
+}
+
+TEST_CASE("bloom_filter: basic union", "[bloom_filter]") {
+  const uint64_t num_bits = 12288;
+  const uint16_t num_hashes = 4;
+
+  auto bf1 = bloom_filter::builder::create_by_size(num_bits, num_hashes);
+  auto bf2 = bloom_filter::builder::create_by_size(num_bits, num_hashes, bf1.get_seed());
+
+  const uint64_t n = 1000;
+  const uint32_t max_item = 3 * n / 2 - 1;
+  for (uint64_t i = 0; i < n; ++i) {
+    bf1.query_and_update(i);
+    bf2.update(n / 2 + i);
+  }
+
+  bf1.union_with(bf2);
+  for (uint64_t i = 0; i < max_item; ++i) {
+    REQUIRE(bf1.query(i));
+  }
+
+  uint32_t num_found = 0;
+  for (uint64_t i = max_item; i < num_bits; ++i) {
+    if (bf1.query(i)) {
+      ++num_found;
+    }
+  }
+  REQUIRE(num_found < num_bits / 10); // not being super strict
+}
+
+TEST_CASE("bloom_filter: basic intersection", "[bloom_filter]") {
+  const uint64_t num_bits = 8192;
+  const uint16_t num_hahes = 5;
+
+  auto bf1 = bloom_filter::builder::create_by_size(num_bits, num_hahes);
+  auto bf2 = bloom_filter::builder::create_by_size(num_bits, num_hahes, bf1.get_seed());
+
+  const uint64_t n = 1024;
+  const uint32_t max_item = 3 * n / 2 - 1;
+  for (uint64_t i = 0; i < n; ++i) {
+    bf1.update(i);
+    bf2.update(n / 2 + i);
+  }
+
+  bf1.intersect(bf2);
+  // overlap bit should all be set
+  for (uint64_t i = n / 2; i < n; ++i) {
+    REQUIRE(bf1.query(i));
+  }
+
+  uint32_t num_found = 0;
+  for (uint64_t i = 0; i < n / 2; ++i) {
+    if (bf1.query(i)) {
+      ++num_found;
+    }
+  }
+  for (uint64_t i = max_item; i < num_bits; ++i) {
+    if (bf1.query(i)) {
+      ++num_found;
+    }
+  }
+
+  REQUIRE(num_found < num_bits / 10); // not being super strict
+}
+
+TEST_CASE("bloom_filter: empty serialization", "[bloom_filter]") {
+  const uint64_t num_bits = 32769;
+  const uint16_t num_hashes = 7;
+
+  auto bf = bloom_filter::builder::create_by_size(num_bits, num_hashes);
+  auto bytes = bf.serialize();
+  REQUIRE(bytes.size() == bf.get_serialized_size_bytes());
+
+  auto bf_bytes = bloom_filter::deserialize(bytes.data(), bytes.size());
+  REQUIRE(bf.get_capacity() == bf_bytes.get_capacity());
+  REQUIRE(bf.get_seed() == bf_bytes.get_seed());
+  REQUIRE(bf.get_num_hashes() == bf_bytes.get_num_hashes());
+  REQUIRE(bf_bytes.is_empty());
+
+  std::stringstream ss(std::ios::in | std::ios::out | std::ios::binary);
+  bf.serialize(ss);
+  auto bf_stream = bloom_filter::deserialize(ss);
+  REQUIRE(bf.get_capacity() == bf_stream.get_capacity());
+  REQUIRE(bf.get_seed() == bf_stream.get_seed());
+  REQUIRE(bf.get_num_hashes() == bf_stream.get_num_hashes());
+  REQUIRE(bf_stream.is_empty());
+
+  // read-only wrap should work
+  auto bf_wrap = bloom_filter::wrap(bytes.data(), bytes.size());
+  REQUIRE(bf.get_capacity() == bf_wrap.get_capacity());
+  REQUIRE(bf.get_seed() == bf_wrap.get_seed());
+  REQUIRE(bf.get_num_hashes() == bf_wrap.get_num_hashes());
+  REQUIRE(bf_wrap.is_empty());
+
+  // writable wrap should not
+  REQUIRE_THROWS_AS(bloom_filter::writable_wrap(bytes.data(), bytes.size()), std::invalid_argument);
+}
+
+TEST_CASE("bloom_filter: non-empty serialization", "[bloom_filter]") {
+  const uint64_t num_bits = 32768;
+  const uint16_t num_hashes = 5;
+
+  auto bf = bloom_filter::builder::create_by_size(num_bits, num_hashes);
+  const uint64_t n = 1000;
+  for (uint64_t i = 0; i < n; ++i) {
+    bf.update(0.5 + i); // testing floats
+  }
+
+  // test more items without updating, assuming some false positives
+  // so we can check that we get the same number of false positives
+  // with the same query items
+  uint64_t fp_count = 0;
+  for (uint64_t i = n; i < num_bits; ++i) {
+    fp_count += bf.query(0.5 + i) ? 1 : 0;
+  }
+
+  auto bytes = bf.serialize();
+  REQUIRE(bytes.size() == bf.get_serialized_size_bytes());
+
+  auto bf_bytes = bloom_filter::deserialize(bytes.data(), bytes.size());
+  REQUIRE(bf.get_capacity() == bf_bytes.get_capacity());
+  REQUIRE(bf.get_seed() == bf_bytes.get_seed());
+  REQUIRE(bf.get_num_hashes() == bf_bytes.get_num_hashes());
+  REQUIRE(!bf_bytes.is_empty());
+  REQUIRE(bf.is_memory_owned());
+  uint64_t fp_count_bytes = 0;
+  for (uint64_t i = 0; i < num_bits; ++i) {
+    bool val = bf_bytes.query(0.5 + i);
+    if (i < n)
+      REQUIRE(val);
+    else if (val)
+      ++fp_count_bytes;
+  }
+  REQUIRE(fp_count_bytes == fp_count);
+
+  std::stringstream ss(std::ios::in | std::ios::out | std::ios::binary);
+  bf.serialize(ss);
+  auto bf_stream = bloom_filter::deserialize(ss);
+  REQUIRE(bf.get_capacity() == bf_stream.get_capacity());
+  REQUIRE(bf.get_seed() == bf_stream.get_seed());
+  REQUIRE(bf.get_num_hashes() == bf_stream.get_num_hashes());
+  REQUIRE(!bf_stream.is_empty());
+  REQUIRE(bf_stream.is_memory_owned());
+  uint64_t fp_count_stream = 0;
+  for (uint64_t i = 0; i < num_bits; ++i) {
+    bool val = bf_stream.query(0.5 + i);
+    if (i < n)
+      REQUIRE(val);
+    else if (val)
+      ++fp_count_stream;
+  }
+  REQUIRE(fp_count_stream == fp_count);
+
+  // read-only wrap
+  auto bf_wrap = bloom_filter::wrap(bytes.data(), bytes.size());
+  REQUIRE(bf.get_capacity() == bf_wrap.get_capacity());
+  REQUIRE(bf.get_seed() == bf_wrap.get_seed());
+  REQUIRE(bf.get_num_hashes() == bf_wrap.get_num_hashes());
+  REQUIRE(!bf_wrap.is_empty());
+  REQUIRE(!bf_wrap.is_memory_owned());
+  uint64_t fp_count_wrap = 0;
+  for (uint64_t i = 0; i < num_bits; ++i) {
+    bool val = bf_wrap.query(0.5 + i);
+    if (i < n)
+      REQUIRE(val);
+    else if (val)
+      ++fp_count_wrap;
+  }
+  REQUIRE(fp_count_wrap == fp_count);
+  REQUIRE_THROWS_AS(bf_wrap.update(-1.0), std::logic_error);
+  REQUIRE_THROWS_AS(bf_wrap.query_and_update(-2.0), std::logic_error);
+  REQUIRE_THROWS_AS(bf_wrap.reset(), std::logic_error);
+
+  // writable wrap
+  auto bf_writable = bloom_filter::writable_wrap(bytes.data(), bytes.size());
+  REQUIRE(bf.get_capacity() == bf_writable.get_capacity());
+  REQUIRE(bf.get_seed() == bf_writable.get_seed());
+  REQUIRE(bf.get_num_hashes() == bf_writable.get_num_hashes());
+  REQUIRE(!bf_writable.is_empty());
+  REQUIRE(!bf_writable.is_memory_owned());
+  uint64_t fp_count_writable = 0;
+  for (uint64_t i = 0; i < num_bits; ++i) {
+    bool val = bf_writable.query(0.5 + i);
+    if (i < n)
+      REQUIRE(val);
+    else if (val)
+      ++fp_count_writable;
+  }
+  REQUIRE(fp_count_writable == fp_count);
+
+  REQUIRE(!bf_writable.query(-1.0));
+  bf_writable.update(-1.0);
+  REQUIRE(bf_writable.query(-1.0));
+
+  // not good memory management to do this, but because we wrapped the same bytes as both
+  // read-only adn writable, that update should ahve changed the read-only version, too
+  REQUIRE(bf_wrap.query(-1.0));
+}
+
+} // namespace datasketches

data/vendor/datasketches-cpp/tdigest/include/tdigest.hpp

@@ -89,6 +89,7 @@ public:
   using vector_t = std::vector<T, Allocator>;
   using vector_centroid = std::vector<centroid, typename std::allocator_traits<Allocator>::template rebind_alloc<centroid>>;
   using vector_bytes = std::vector<uint8_t, typename std::allocator_traits<Allocator>::template rebind_alloc<uint8_t>>;
+  using vector_double = std::vector<double, typename std::allocator_traits<Allocator>::template rebind_alloc<double>>;
 
   struct centroid_cmp {
     centroid_cmp() {}
@@ -115,7 +116,7 @@ public:
    * Merge the given t-Digest into this one
    * @param other t-Digest to merge
    */
-  void merge(tdigest& other);
+  void merge(const tdigest& other);
 
   /**
    * Process buffered values and merge centroids if needed
@@ -142,8 +143,17 @@ public:
    */
   uint64_t get_total_weight() const;
 
+  /**
+   * Returns an instance of the allocator for this t-Digest.
+   * @return allocator
+   */
+  Allocator get_allocator() const;
+
   /**
    * Compute approximate normalized rank of the given value.
+   *
+   * <p>If the sketch is empty this throws std::runtime_error.
+   *
    * @param value to be ranked
    * @return normalized rank (from 0 to 1 inclusive)
    */
@@ -151,11 +161,49 @@ public:
 
   /**
    * Compute approximate quantile value corresponding to the given normalized rank
+   *
+   * <p>If the sketch is empty this throws std::runtime_error.
+   *
    * @param rank normalized rank (from 0 to 1 inclusive)
    * @return quantile value corresponding to the given rank
    */
   T get_quantile(double rank) const;
 
+  /**
+   * Returns an approximation to the Probability Mass Function (PMF) of the input stream
+   * given a set of split points.
+   *
+   * <p>If the sketch is empty this throws std::runtime_error.
+   *
+   * @param split_points an array of <i>m</i> unique, monotonically increasing values
+   * that divide the input domain into <i>m+1</i> consecutive disjoint intervals (bins).
+   *
+   * @param size the number of split points in the array
+   *
+   * @return an array of m+1 doubles each of which is an approximation
+   * to the fraction of the input stream values (the mass) that fall into one of those intervals.
+   */
+  vector_double get_PMF(const T* split_points, uint32_t size) const;
+
+  /**
+   * Returns an approximation to the Cumulative Distribution Function (CDF), which is the
+   * cumulative analog of the PMF, of the input stream given a set of split points.
+   *
+   * <p>If the sketch is empty this throws std::runtime_error.
+   *
+   * @param split_points an array of <i>m</i> unique, monotonically increasing values
+   * that divide the input domain into <i>m+1</i> consecutive disjoint intervals.
+   *
+   * @param size the number of split points in the array
+   *
+   * @return an array of m+1 doubles, which are a consecutive approximation to the CDF
+   * of the input stream given the split_points. The value at array position j of the returned
+   * CDF array is the sum of the returned values in positions 0 through j of the returned PMF
+   * array. This can be viewed as array of ranks of the given split points plus one more value
+   * that is always 1.
+   */
+  vector_double get_CDF(const T* split_points, uint32_t size) const;
+
   /**
    * @return parameter k (compression) that was used to configure this t-Digest
    */
@@ -245,6 +293,8 @@ private:
   // for compatibility with format of the reference implementation
   static tdigest deserialize_compat(std::istream& is, const Allocator& allocator = Allocator());
   static tdigest deserialize_compat(const void* bytes, size_t size, const Allocator& allocator = Allocator());
+
+  static inline void check_split_points(const T* values, uint32_t size);
 };
 
 } /* namespace datasketches */

data/vendor/datasketches-cpp/tdigest/include/tdigest_impl.hpp

@@ -20,6 +20,7 @@
 #ifndef _TDIGEST_IMPL_HPP_
 #define _TDIGEST_IMPL_HPP_
 
+#include <algorithm>
 #include <cmath>
 #include <sstream>
 
@@ -43,7 +44,7 @@ void tdigest<T, A>::update(T value) {
 }
 
 template<typename T, typename A>
-void tdigest<T, A>::merge(tdigest& other) {
+void tdigest<T, A>::merge(const tdigest& other) {
   if (other.is_empty()) return;
   vector_centroid tmp(buffer_.get_allocator());
   tmp.reserve(buffer_.size() + centroids_.size() + other.buffer_.size() + other.centroids_.size());
@@ -84,6 +85,11 @@ uint64_t tdigest<T, A>::get_total_weight() const {
   return centroids_weight_ + buffer_.size();
 }
 
+template<typename T, typename A>
+A tdigest<T, A>::get_allocator() const {
+  return buffer_.get_allocator();
+}
+
 template<typename T, typename A>
 double tdigest<T, A>::get_rank(T value) const {
   if (is_empty()) throw std::runtime_error("operation is undefined for an empty sketch");
@@ -190,6 +196,25 @@ T tdigest<T, A>::get_quantile(double rank) const {
   return weighted_average(centroids_.back().get_weight(), w1, max_, w2);
 }
 
+template<typename T, typename A>
+auto tdigest<T, A>::get_PMF(const T* split_points, uint32_t size) const -> vector_double {
+  auto buckets = get_CDF(split_points, size);
+  for (uint32_t i = size; i > 0; --i) {
+    buckets[i] -= buckets[i - 1];
+  }
+  return buckets;
+}
+
+template<typename T, typename A>
+auto tdigest<T, A>::get_CDF(const T* split_points, uint32_t size) const -> vector_double {
+  check_split_points(split_points, size);
+  vector_double ranks(get_allocator());
+  ranks.reserve(size + 1);
+  for (uint32_t i = 0; i < size; ++i) ranks.push_back(get_rank(split_points[i]));
+  ranks.push_back(1);
+  return ranks;
+}
+
 template<typename T, typename A>
 uint16_t tdigest<T, A>::get_k() const {
   return k_;
@@ -590,6 +615,18 @@ buffer_(std::move(buffer))
   buffer_.reserve(centroids_capacity_ * BUFFER_MULTIPLIER);
 }
 
+template<typename T, typename A>
+void tdigest<T, A>::check_split_points(const T* values, uint32_t size) {
+  for (uint32_t i = 0; i < size ; i++) {
+    if (std::isnan(values[i])) {
+      throw std::invalid_argument("Values must not be NaN");
+    }
+    if ((i < (size - 1)) && !(values[i] < values[i + 1])) {
+      throw std::invalid_argument("Values must be unique and monotonically increasing");
+    }
+  }
+}
+
 } /* namespace datasketches */
 
 #endif // _TDIGEST_IMPL_HPP_

data/vendor/datasketches-cpp/tdigest/test/tdigest_test.cpp

@@ -35,6 +35,9 @@ TEST_CASE("empty", "[tdigest]") {
   REQUIRE_THROWS_AS(td.get_max_value(), std::runtime_error);
   REQUIRE_THROWS_AS(td.get_rank(0), std::runtime_error);
   REQUIRE_THROWS_AS(td.get_quantile(0.5), std::runtime_error);
+  const double split_points[1] {0};
+  REQUIRE_THROWS_AS(td.get_PMF(split_points, 1), std::runtime_error);
+  REQUIRE_THROWS_AS(td.get_CDF(split_points, 1), std::runtime_error);
 }
 
 TEST_CASE("one value", "[tdigest]") {
@@ -56,9 +59,6 @@ TEST_CASE("many values", "[tdigest]") {
   const size_t n = 10000;
   tdigest_double td;
   for (size_t i = 0; i < n; ++i) td.update(i);
-//  std::cout << td.to_string(true);
-//  td.compress();
-//  std::cout << td.to_string(true);
   REQUIRE_FALSE(td.is_empty());
   REQUIRE(td.get_total_weight() == n);
   REQUIRE(td.get_min_value() == 0);
@@ -73,6 +73,15 @@ TEST_CASE("many values", "[tdigest]") {
   REQUIRE(td.get_quantile(0.9) == Approx(n * 0.9).epsilon(0.01));
   REQUIRE(td.get_quantile(0.95) == Approx(n * 0.95).epsilon(0.01));
   REQUIRE(td.get_quantile(1) == n - 1);
+  const double split_points[1] {n / 2};
+  const auto pmf = td.get_PMF(split_points, 1);
+  REQUIRE(pmf.size() == 2);
+  REQUIRE(pmf[0] == Approx(0.5).margin(0.0001));
+  REQUIRE(pmf[1] == Approx(0.5).margin(0.0001));
+  const auto cdf = td.get_CDF(split_points, 1);
+  REQUIRE(cdf.size() == 2);
+  REQUIRE(cdf[0] == Approx(0.5).margin(0.0001));
+  REQUIRE(cdf[1] == 1);
 }
 
 TEST_CASE("rank - two values", "[tdigest]") {

data/vendor/datasketches-cpp/theta/include/bit_packing.hpp

@@ -329,7 +329,7 @@ static inline void pack_bits_13(const uint64_t* values, uint8_t* ptr) {
 
   *ptr++ = static_cast<uint8_t>(values[3] >> 4);
 
-  *ptr = static_cast<uint8_t>(values[3] >> 4);
+  *ptr = static_cast<uint8_t>(values[3] << 4);
   *ptr++ |= static_cast<uint8_t>(values[4] >> 9);
 
   *ptr++ = static_cast<uint8_t>(values[4] >> 1);
@@ -4227,7 +4227,7 @@ static inline void unpack_bits_33(uint64_t* values, const uint8_t* ptr) {
   values[6] |= *ptr >> 1;
 
   values[7] = static_cast<uint64_t>(*ptr++ & 1) << 32;
-  values[7] |= *ptr++ << 24;
+  values[7] |= static_cast<uint64_t>(*ptr++) << 24;
   values[7] |= *ptr++ << 16;
   values[7] |= *ptr++ << 8;
   values[7] |= *ptr;
@@ -4296,7 +4296,7 @@ static inline void unpack_bits_35(uint64_t* values, const uint8_t* ptr) {
   values[1] |= *ptr++ << 6;
   values[1] |= *ptr >> 2;
 
-  values[2] = static_cast<uint64_t>(*ptr++ & 2) << 33;
+  values[2] = static_cast<uint64_t>(*ptr++ & 3) << 33;
   values[2] |= static_cast<uint64_t>(*ptr++) << 25;
   values[2] |= *ptr++ << 17;
   values[2] |= *ptr++ << 9;
@@ -6201,7 +6201,7 @@ static inline void pack_bits_block8(const uint64_t* values, uint8_t* ptr, uint8_
     case 61: pack_bits_61(values, ptr); break;
     case 62: pack_bits_62(values, ptr); break;
     case 63: pack_bits_63(values, ptr); break;
-    default: throw std::logic_error("wrong number of bits " + std::to_string(bits));
+    default: throw std::logic_error("wrong number of bits in pack_bits_block8: " + std::to_string(bits));
   }
 }
 
@@ -6270,7 +6270,7 @@ static inline void unpack_bits_block8(uint64_t* values, const uint8_t* ptr, uint
     case 61: unpack_bits_61(values, ptr); break;
     case 62: unpack_bits_62(values, ptr); break;
     case 63: unpack_bits_63(values, ptr); break;
-    default: throw std::logic_error("wrong number of bits " + std::to_string(bits));
+    default: throw std::logic_error("wrong number of bits in unpack_bits_block8: " + std::to_string(bits));
   }
 }