我试图加速对3D数组沿Z轴的平均值计算。我阅读了Cython文档以添加类型、内存视图等来完成此任务。然而,当我比较基于NumPy和基于Cython语法及编译生成.so文件的函数时,前者胜过后者。请问我的代码中是否有哪些步骤或类型声明存在问题或遗漏?
这是我的NumPy版本:python_mean.py
import numpy as np
def mean_py(array):
x = array.shape[1]
y = array.shape[2]
values = []
for i in range(x):
for j in range(y):
values.append((np.mean(array[:, i, j])))
values = np.array([values])
values = values.reshape(500,500)
return values
这是我的 cython_mean.pyx 文件。
%%cython
from cython import wraparound, boundscheck
import numpy as np
cimport numpy as np
DTYPE = np.double
@boundscheck(False)
@wraparound(False)
def cy_mean(double[:,:,:] array):
cdef Py_ssize_t x_max = array.shape[1]
cdef Py_ssize_t y_max = array.shape[2]
cdef double[:,:] result = np.zeros([x_max, y_max], dtype = DTYPE)
cdef double[:,:] result_view = result
cdef Py_ssize_t i,j
cdef double mean
cdef list values
for i in range(x_max):
for j in range(y_max):
mean = np.mean(array[:,i,j])
result_view[i,j] = mean
return result
当我导入两个函数并开始对一个3D numpy数组进行计算时,我得到了以下结果:
import numpy as np
a = np.random.randn(250_000)
b = np.random.randn(250_000)
c = np.random.randn(250_000)
array = np.vstack((a,b,c)).reshape(3, 500, 500)
import mean_py
from mean_py import mean_py
%timeit mean_py(array)
4.82 s ± 84.7 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)
import cython_mean
from cython_mean import cy_mean
7.3 s ± 499 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)
为什么Cython代码性能如此低?感谢您的帮助。