Java 跳跃列表生成器使用哪个 C++ 随机数分布？

Question

JavaConcurrentSkipListMapclass包含一个方法randomLevel，根据多处理器编程艺术[=25=，输出以下内容]:

The randomLevel() method is designed based on empirical measurements to maintain the skiplist property. For example, in the java.util.concurrent package, for a maximal SkipList level of 31, randomLevel() returns 0 with probability 3/4, i with probability 2^(−(i+2)) for i ∈ [1, 30], and 31 with probability 2^−32.

这个看起来像geometric distribution，但不完全是。有什么方法可以根据提供的随机分布巧妙地定义它，还是我必须自己进行操作，例如：

inline unsigned randomLevel() { auto randNum = distribution.operator()(engine); // distribution is std::uniform_int_distribution<> unsigned two__30{0x4000'0000}; if (randNum == 0) return 31; // p(level == 31) = 2**-31 else if (randNum >= two__30) return 0; // p(level = 0) = 0.75 else return 30 - static_cast<unsigned>(log2(randNum)); // p(level = i) = 2**-(i+2) }

Answer 1

This looks like a geometric distribution, but not quite.

你是对的，但问题只是 0 的概率。请注意，您可以通过将前两个值合并为一个来使用 std::geometric_distribution。

class RandomLevel
{
    std::geometric_distribution<unsigned> distribution;
    std::mt19937 gen{std::random_device{}()};
public:
    unsigned operator()() {
        auto result = distribution(gen);
        return result > 1u : result - 1u : 0u;
    }
}

Java 跳跃列表生成器使用哪个 C++ 随机数分布？

Which C++ random number distribution to use for the Java skip list generator?

c++

random