数组中的 Python Numba 值

Question

我正在尝试检查一个数字是否在 int8 的 NumPy 数组中。我试过这个，但它不起作用。

from numba import njit
import numpy as np

@njit
def c(b):
    return 9 in b

a = np.array((9, 10, 11), 'int8')
print(c(a))

我得到的错误是

Invalid use of Function(<built-in function contains>) with argument(s) of type(s): (array(int8, 1d, C), Literal[int](9))
 * parameterized
In definition 0:
    All templates rejected with literals.
In definition 1:
    All templates rejected without literals.
In definition 2:
    All templates rejected with literals.
In definition 3:
    All templates rejected without literals.
In definition 4:
    All templates rejected with literals.
In definition 5:
    All templates rejected without literals.
This error is usually caused by passing an argument of a type that is unsupported by the named function.
[1] During: typing of intrinsic-call at .\emptyList.py (6)

如何在保持性能的同时解决这个问题？将检查数组是否有两个值，1 和 -1，长度为 32 项。它们没有排序。

Answer 1

检查两个值是否在数组中

为了仅检查两个值是否出现在数组中，我建议使用简单的暴力算法。

代码

import numba as nb
import numpy as np

@nb.njit(fastmath=True)
def isin(b):
  for i in range(b.shape[0]):
    res=False
    if (b[i]==-1):
      res=True
    if (b[i]==1):
      res=True
  return res

#Parallelized call to isin if the data is an array of shape (n,m)
@nb.njit(fastmath=True,parallel=True)
def isin_arr(b):
  res=np.empty(b.shape[0],dtype=nb.boolean)
  for i in nb.prange(b.shape[0]):
    res[i]=isin(b[i,:])

  return res

性能

#Create some data (320MB)
A=(np.random.randn(10000000,32)-0.5)*5
A=A.astype(np.int8)
res=isin_arr(A) 11ms per call

因此，通过这种方法，我获得了大约 29GB/s 的吞吐量，这与内存带宽相差不远。您还可以尝试减少 Testdatasize，使其适合 L3 缓存以避免内存带宽限制。使用 3.2 MB 测试数据时，我获得了 100 GB/s 的吞吐量(远远超出了我的内存带宽)，这清楚地表明此实现的内存带宽有限。