在 Python 上使用 Numba 的 Numpy 均值和标准差

Question

I am trying to use Numpy's mean and standard deviation functions insinde a function and they don't seem to be compatible with Numba, although Numba documentation states them as compatible .

My code is the following:

import numpy as np
import numba


a = [1, 2, 3, 4, 5, 6]

# @numba.jit(nopython=True, parallel=True)
def nmeanstd(a, n):
    b = []; c = []
    for i in range(n):
        b.append(np.mean(a))
        c.append(np.std(a))
    
    return b, c

mean, std = nmeanstd(a, 10)

The output when looking at mean and std is the expected:

mean
Out[31]: [3.5, 3.5, 3.5, 3.5, 3.5, 3.5, 3.5, 3.5, 3.5, 3.5]

std
Out[32]: 
[1.707825127659933,
 1.707825127659933,
 1.707825127659933,
 1.707825127659933,
 1.707825127659933,
 1.707825127659933,
 1.707825127659933,
 1.707825127659933,
 1.707825127659933,
 1.707825127659933]

But I don't know why, when I uncomment the @numba.jit function the following message appears:

TypingError: No implementation of function Function(<function mean at 0x11a0e6e50>) found for signature:
 
mean(reflected list(int64)<iv=None>)
 
There are 2 candidate implementations:
      - Of which 2 did not match due to:
      Overload of function 'mean': File: numba/core/typing/npydecl.py: Line 378.
        With argument(s): '(reflected list(int64)<iv=None>)':
       No match.

During: resolving callee type: Function(<function mean at 0x11a0e6e50>)

And the same for std if I comment the line in which I compute the mean. What is happening? I though they would be running with numba correctly. Do you know any way of computing the mean and the standard deviation using Numba?

Answer 1

The error message shows that Numba does not know how to compute the mean of a list . Your code works fine (with @jit ) if the input list is first converted to a numpy array:

mean, std = nmeanstd(np.array(a), 10)

Answer 2

The documentation shows Numpy arrays work directly with Numba and are very efficient.

The code works if you convert "a" into a numpy array mean, std = nmeanstd(np.array(a), 10)