多线程生成

四个核心分布( random , standard_normal , standard_exponential 和 standard_gamma )都允许使用 out 关键字参数填充现有数组.现有数组需要是连续的且行为良好(可写且对齐).在正常情况下,使用常见构造函数(例如 numpy.empty )创建的数组将满足这些要求.

此示例利用 concurrent.futures 使用多个线程填充数组.线程是长期存在的,因此重复调用不需要线程创建的任何额外开销.

生成的随机数是可重现的,因为在线程数不变的情况下,相同的种子将产生相同的输出.

from numpy.random import default_rng, SeedSequence
import multiprocessing
import concurrent.futures
import numpy as np

class MultithreadedRNG:
    def __init__(self, n, seed=None, threads=None):
        if threads is None:
            threads = multiprocessing.cpu_count()
        self.threads = threads

        seq = SeedSequence(seed)
        self._random_generators = [default_rng(s)
                                   for s in seq.spawn(threads)]

        self.n = n
        self.executor = concurrent.futures.ThreadPoolExecutor(threads)
        self.values = np.empty(n)
        self.step = np.ceil(n / threads).astype(np.int_)

    def fill(self):
        def _fill(random_state, out, first, last):
            random_state.standard_normal(out=out[first:last])

        futures = {}
        for i in range(self.threads):
            args = (_fill,
                    self._random_generators[i],
                    self.values,
                    i * self.step,
                    (i + 1) * self.step)
            futures[self.executor.submit(*args)] = i
        concurrent.futures.wait(futures)

    def __del__(self):
        self.executor.shutdown(False)

多线程随机数生成器可用于填充数组. values 属性显示的是填充前的零值和填充后的随机值.

In [2]: mrng = MultithreadedRNG(10000000, seed=12345)
   ...: print(mrng.values[-1])
Out[2]: 0.0

In [3]: mrng.fill()
   ...: print(mrng.values[-1])
Out[3]: 2.4545724517479104

可以使用多个线程生成所需的时间与使用单个线程生成所需的时间进行比较.

In [4]: print(mrng.threads)
   ...: %timeit mrng.fill()

Out[4]: 4
   ...: 32.8 ms ± 2.71 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)

单线程调用直接使用 BitGenerator.

In [5]: values = np.empty(10000000)
   ...: rg = default_rng()
   ...: %timeit rg.standard_normal(out=values)

Out[5]: 99.6 ms ± 222 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)

即使对于仅是中等大小的数组,增益也很显著,并且缩放是合理的.与由于数组创建开销而不使用现有数组的调用相比,增益甚至更大.

In [6]: rg = default_rng()
   ...: %timeit rg.standard_normal(10000000)

Out[6]: 125 ms ± 309 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)

请注意,如果用户未设置 threads ,它将由 multiprocessing.cpu_count() 确定.

In [7]: # simulate the behavior for `threads=None`, if the machine had only one thread
   ...: mrng = MultithreadedRNG(10000000, seed=12345, threads=1)
   ...: print(mrng.values[-1])
Out[7]: 1.1800150052158556