python numpy numba 计算速度对比

其他 2019-01-07 22:05:03 阅读次数: 0
版权声明：版权声明：本文为博主原创文章，未经博主允许不得转载。 https://blog.csdn.net/DarrenXf/article/details/85863796 
#!/usr/bin/python
# -*- coding: utf-8 -*-
#####################################
# File name : test.py
# Create date : 2019-01-05 17:11
# Modified date : 2019-01-05 17:59
# Author : DARREN
# Describe : not set
# Email : [email protected]
#####################################
from __future__ import division
from __future__ import print_function

import numpy as np
from timeit import Timer
from numba import jit

def python_list(n):
    a = range(n)
    b = range(n)
    c = []
    for i in range(len(a)):
        a[i] = i ** 2
        b[i] = i ** 3
        c.append(a[i] + b[i])
    return c

@jit
def python_list_with_numba(n):
#   a = range(n)
#   b = range(n)
#   c = []
#   for i in range(len(a)):
#       a[i] = i ** 2
#       b[i] = i ** 3
#       c.append(a[i] + b[i])
#   return c
    return python_list(n)

def numpy_vector(n):
    x = np.arange(n)
    y = np.arange(n)
    z = x**2 + y**3
    return z

@jit
def numpy_vector_with_numba(n):
#   x = np.arange(n)
#   y = np.arange(n)
#   z = x**2 + y**3
#   return z
    return numpy_vector(n)


if __name__ == '__main__':
    t1 = Timer('python_list(1000000)','from __main__ import python_list')
    t2 = Timer('python_list_with_numba(1000000)','from __main__ import python_list_with_numba')
    t3 = Timer('python_list_with_numba(1000000)','from __main__ import python_list_with_numba')
    t4 = Timer('numpy_vector(1000000)','from __main__ import numpy_vector')
    t5 = Timer('numpy_vector_with_numba(1000000)','from __main__ import numpy_vector_with_numba')
    t6 = Timer('numpy_vector_with_numba(1000000)','from __main__ import numpy_vector_with_numba')

    take_time1 = t1.timeit(50)
    take_time2 = t2.timeit(50)
    take_time3 = t3.timeit(50)
    take_time4 = t4.timeit(50)
    take_time5 = t5.timeit(50)
    take_time6 = t6.timeit(50)

    print("python list:%s" % take_time1)
    print("python list with numba comp:%s" % take_time2)
    print("python list with numba after comp:%s" % take_time3)
    print("numpy vector:%s" % take_time4)
    print("numpy vector with numba comp:%s" % take_time5)
    print("numpy vector with numba after comp:%s" % take_time6)

output

python list:8.37137699127
python list with numba comp:8.26541399956
python list with numba after comp:8.33281493187
numpy vector:0.529839992523
numpy vector with numba comp:0.539204835892
numpy vector with numba after comp:0.524770021439

好像调用函数对与 numba来说是有有影响的.调用函数的话numba 就起不到加速的效果了.下面去掉函数调用

#!/usr/bin/python
# -*- coding: utf-8 -*-
#####################################
# File name : test.py
# Create date : 2019-01-05 17:11
# Modified date : 2019-01-05 18:05
# Author : DARREN
# Describe : not set
# Email : [email protected]
#####################################
from __future__ import division
from __future__ import print_function

import numpy as np
from timeit import Timer
from numba import jit

def python_list(n):
    a = range(n)
    b = range(n)
    c = []
    for i in range(len(a)):
        a[i] = i ** 2
        b[i] = i ** 3
        c.append(a[i] + b[i])
    return c

@jit
def python_list_with_numba(n):
    a = range(n)
    b = range(n)
    c = []
    for i in range(len(a)):
        a[i] = i ** 2
        b[i] = i ** 3
        c.append(a[i] + b[i])
    return c

def numpy_vector(n):
    x = np.arange(n)
    y = np.arange(n)
    z = x**2 + y**3
    return z

@jit
def numpy_vector_with_numba(n):
    x = np.arange(n)
    y = np.arange(n)
    z = x**2 + y**3
    return z


if __name__ == '__main__':
    t1 = Timer('python_list(1000000)','from __main__ import python_list')
    t2 = Timer('python_list_with_numba(1000000)','from __main__ import python_list_with_numba')
    t3 = Timer('python_list_with_numba(1000000)','from __main__ import python_list_with_numba')
    t4 = Timer('numpy_vector(1000000)','from __main__ import numpy_vector')
    t5 = Timer('numpy_vector_with_numba(1000000)','from __main__ import numpy_vector_with_numba')
    t6 = Timer('numpy_vector_with_numba(1000000)','from __main__ import numpy_vector_with_numba')

    take_time1 = t1.timeit(50)
    take_time2 = t2.timeit(50)
    take_time3 = t3.timeit(50)
    take_time4 = t4.timeit(50)
    take_time5 = t5.timeit(50)
    take_time6 = t6.timeit(50)

    print("python list:%s" % take_time1)
    print("python list with numba comp:%s" % take_time2)
    print("python list with numba after comp:%s" % take_time3)
    print("numpy vector:%s" % take_time4)
    print("numpy vector with numba comp:%s" % take_time5)
    print("numpy vector with numba after comp:%s" % take_time6)

output

python list:8.16710710526
python list with numba comp:9.53106093407
python list with numba after comp:9.40612602234
numpy vector:0.188210964203
numpy vector with numba comp:0.314162015915
numpy vector with numba after comp:0.0830471515656

numpy在计算上比python 要快.加上numba后numpy似乎可以提高1倍的速度.
numba编译的那次速度是比较慢的.

#!/usr/bin/python
# -*- coding: utf-8 -*-
#####################################
# File name : test.py
# Create date : 2019-01-05 17:11
# Modified date : 2019-01-05 21:27
# Author : DARREN
# Describe : not set
# Email : [email protected]
#####################################
from __future__ import division
from __future__ import print_function

import numpy as np
from timeit import Timer
from numba import jit
from numba import njit

def python_list(n):
    a = range(n)
    b = range(n)
    c = []
    for i in range(len(a)):
        a[i] = i ** 2
        b[i] = i ** 3
        c.append(a[i] + b[i])
    return c

@jit
def python_list_with_numba(n):
    a = range(n)
    b = range(n)
    c = []
    for i in range(len(a)):
        a[i] = i ** 2
        b[i] = i ** 3
        c.append(a[i] + b[i])
    return c

def numpy_vector(n):
    x = np.arange(n)
    y = np.arange(n)
    z = x**2 + y**3
    return z

@jit
def numpy_vector_with_numba(n):
    x = np.arange(n)
    y = np.arange(n)
    z = x**2 + y**3
    return z

@njit(fastmath=True)
def numpy_vector_with_numba_fast(n):
    x = np.arange(n)
    y = np.arange(n)
    z = x**2 + y**3
    return z

@njit(parallel=True)
def numpy_vector_with_numba_parallel(n):
    x = np.arange(n)
    y = np.arange(n)
    z = x**2 + y**3
    return z

@njit(fastmath=True, parallel=True)
def numpy_vector_with_numba_fast_parallel(n):
    x = np.arange(n)
    y = np.arange(n)
    z = x**2 + y**3
    return z


if __name__ == '__main__':
    t1 = Timer('python_list(1000000)','from __main__ import python_list')
    t2 = Timer('python_list_with_numba(1000000)','from __main__ import python_list_with_numba')
    t3 = Timer('python_list_with_numba(1000000)','from __main__ import python_list_with_numba')
    t4 = Timer('numpy_vector(1000000)','from __main__ import numpy_vector')
    t5 = Timer('numpy_vector_with_numba(1000000)','from __main__ import numpy_vector_with_numba')
    t6 = Timer('numpy_vector_with_numba(1000000)','from __main__ import numpy_vector_with_numba')
    t7 = Timer('numpy_vector_with_numba_fast(1000000)','from __main__ import numpy_vector_with_numba_fast')
    t8 = Timer('numpy_vector_with_numba_fast(1000000)','from __main__ import numpy_vector_with_numba_fast')

    t9 = Timer('numpy_vector_with_numba_parallel(1000000)','from __main__ import numpy_vector_with_numba_parallel')
    t10 = Timer('numpy_vector_with_numba_parallel(1000000)','from __main__ import numpy_vector_with_numba_parallel')

    t11 = Timer('numpy_vector_with_numba_fast_parallel(1000000)','from __main__ import numpy_vector_with_numba_fast_parallel')
    t12 = Timer('numpy_vector_with_numba_fast_parallel(1000000)','from __main__ import numpy_vector_with_numba_fast_parallel')

    take_time1 = t1.timeit(50)
    take_time2 = t2.timeit(50)
    take_time3 = t3.timeit(50)
    take_time4 = t4.timeit(50)
    take_time5 = t5.timeit(50)
    take_time6 = t6.timeit(50)
    take_time7 = t7.timeit(50)
    take_time8 = t8.timeit(50)
    take_time9 = t9.timeit(50)
    take_time10 = t10.timeit(50)
    take_time11 = t11.timeit(50)
    take_time12 = t12.timeit(50)

    print("python_list:%s" % take_time1)
    print("python_list_with_numba compile:%s" % take_time2)
    print("python_list_with_numba after compile:%s" % take_time3)
    print("numpy_vector:%s" % take_time4)
    print("numpy_vector_with_numba compile:%s" % take_time5)
    print("numpy_vector_with_numba after compile:%s" % take_time6)
    print("numpy_vector_with_numba_fast compile:%s" % take_time7)
    print("numpy_vector_with_numba_fast after compile:%s" % take_time8)
    print("numpy_vector_with_numba_parallel compile:%s" % take_time9)
    print("numpy_vector_with_numba_parallel after compile:%s" % take_time10)
    print("numpy_vector_with_numba_fast_parallel compile:%s" % take_time11)
    print("numpy_vector_with_numba_fast_parallel after compile:%s" % take_time12)

output

python_list:8.22160506248
python_list_with_numba compile:9.52428817749
python_list_with_numba after compile:9.4302740097
numpy_vector:0.18684220314
numpy_vector_with_numba compile:0.313122034073
numpy_vector_with_numba after compile:0.0834867954254
numpy_vector_with_numba_fast compile:0.166754961014
numpy_vector_with_numba_fast after compile:0.0827031135559
numpy_vector_with_numba_parallel compile:0.353137016296
numpy_vector_with_numba_parallel after compile:0.0367288589478
numpy_vector_with_numba_fast_parallel compile:0.359938144684
numpy_vector_with_numba_fast_parallel after compile:0.0356090068817

加上parallel 能够提速6倍.看起来确实加速了.

下面测试下矩阵相乘 是否能被加速

#!/usr/bin/python
# -*- coding: utf-8 -*-
#####################################
# File name : test.py
# Create date : 2019-01-05 17:11
# Modified date : 2019-01-05 22:22
# Author : DARREN
# Describe : not set
# Email : [email protected]
#####################################
from __future__ import division
from __future__ import print_function

import numpy as np
from timeit import Timer
from numba import jit
from numba import njit

def dot(x,y):
    z = np.dot(x,y)
    return z

@jit
def dot_numba(x,y):
    z = np.dot(x,y)
    return z

@njit(fastmath=True)
def dot_numba_fast(x,y):
    z = np.dot(x,y)
    return z

@njit(parallel=True)
def dot_numba_parallel(x,y):
    z = np.dot(x,y)
    return z

@njit(fastmath=True, parallel=True)
def dot_numba_fast_parallel(x,y):
    z = np.dot(x,y)
    return z

if __name__ == '__main__':

    x = np.random.randn(100,100)
    y = np.random.randn(100,100)

    t1 = Timer('dot(x,y)','from __main__ import dot,x,y')
    t2 = Timer('dot_numba(x,y)','from __main__ import dot_numba,x,y')
    t3 = Timer('dot_numba(x,y)','from __main__ import dot_numba,x,y')
    t4 = Timer('dot_numba_fast(x,y)','from __main__ import dot_numba_fast,x,y')
    t5 = Timer('dot_numba_fast(x,y)','from __main__ import dot_numba_fast,x,y')

    t6 = Timer('dot_numba_parallel(x,y)','from __main__ import dot_numba_parallel,x,y')
    t7 = Timer('dot_numba_parallel(x,y)','from __main__ import dot_numba_parallel,x,y')

    t8 = Timer('dot_numba_fast_parallel(x,y)','from __main__ import dot_numba_fast_parallel,x,y')
    t9 = Timer('dot_numba_fast_parallel(x,y)','from __main__ import dot_numba_fast_parallel,x,y')

    take_time1 = t1.timeit(90)
    take_time2 = t2.timeit(90)
    take_time3 = t3.timeit(90)
    take_time4 = t4.timeit(90)
    take_time5 = t5.timeit(90)
    take_time6 = t6.timeit(90)
    take_time7 = t7.timeit(90)
    take_time8 = t8.timeit(90)
    take_time9 = t9.timeit(90)

    print("dot:%s" % take_time1)
    print("dot_numba compilation:%s" % take_time2)
    print("dot_numba after compilation:%s" % take_time3)
    print("dot_numba_fast compilation:%s" % take_time4)
    print("dot_numba_fast after compilation:%s" % take_time5)
    print("dot_numba_parallel compilation:%s" % take_time6)
    print("dot_numba_parallel after compilation:%s" % take_time7)
    print("dot_numba_fast_parallel compilation:%s" % take_time8)
    print("dot_numba_fast_parallel after compilation:%s" % take_time9)

output

dot:0.0045690536499
dot_numba compilation:0.161776065826
dot_numba after compilation:0.00354194641113
dot_numba_fast compilation:0.0609059333801
dot_numba_fast after compilation:0.00446605682373
dot_numba_parallel compilation:0.0760118961334
dot_numba_parallel after compilation:0.00450396537781
dot_numba_fast_parallel compilation:0.0754170417786
dot_numba_fast_parallel after compilation:0.00403594970703

能够看到编译的时候会更慢,编译后的numba也不比numpy快.
下面把矩阵变大看下效果

#!/usr/bin/python
# -*- coding: utf-8 -*-
#####################################
# File name : test.py
# Create date : 2019-01-05 17:11
# Modified date : 2019-01-05 22:26
# Author : DARREN
# Describe : not set
# Email : [email protected]
#####################################
from __future__ import division
from __future__ import print_function

import numpy as np
from timeit import Timer
from numba import jit
from numba import njit

def dot(x,y):
    z = np.dot(x,y)
    return z

@jit
def dot_numba(x,y):
    z = np.dot(x,y)
    return z

@njit(fastmath=True)
def dot_numba_fast(x,y):
    z = np.dot(x,y)
    return z

@njit(parallel=True)
def dot_numba_parallel(x,y):
    z = np.dot(x,y)
    return z

@njit(fastmath=True, parallel=True)
def dot_numba_fast_parallel(x,y):
    z = np.dot(x,y)
    return z

if __name__ == '__main__':

    x = np.random.randn(1000,1000)
    y = np.random.randn(1000,1000)

    t1 = Timer('dot(x,y)','from __main__ import dot,x,y')
    t2 = Timer('dot_numba(x,y)','from __main__ import dot_numba,x,y')
    t3 = Timer('dot_numba(x,y)','from __main__ import dot_numba,x,y')
    t4 = Timer('dot_numba_fast(x,y)','from __main__ import dot_numba_fast,x,y')
    t5 = Timer('dot_numba_fast(x,y)','from __main__ import dot_numba_fast,x,y')

    t6 = Timer('dot_numba_parallel(x,y)','from __main__ import dot_numba_parallel,x,y')
    t7 = Timer('dot_numba_parallel(x,y)','from __main__ import dot_numba_parallel,x,y')

    t8 = Timer('dot_numba_fast_parallel(x,y)','from __main__ import dot_numba_fast_parallel,x,y')
    t9 = Timer('dot_numba_fast_parallel(x,y)','from __main__ import dot_numba_fast_parallel,x,y')

    take_time1 = t1.timeit(90)
    take_time2 = t2.timeit(90)
    take_time3 = t3.timeit(90)
    take_time4 = t4.timeit(90)
    take_time5 = t5.timeit(90)
    take_time6 = t6.timeit(90)
    take_time7 = t7.timeit(90)
    take_time8 = t8.timeit(90)
    take_time9 = t9.timeit(90)

    print("dot:%s" % take_time1)
    print("dot_numba compilation:%s" % take_time2)
    print("dot_numba after compilation:%s" % take_time3)
    print("dot_numba_fast compilation:%s" % take_time4)
    print("dot_numba_fast after compilation:%s" % take_time5)
    print("dot_numba_parallel compilation:%s" % take_time6)
    print("dot_numba_parallel after compilation:%s" % take_time7)
    print("dot_numba_fast_parallel compilation:%s" % take_time8)
    print("dot_numba_fast_parallel after compilation:%s" % take_time9)

output

dot:1.29017019272
dot_numba compilation:1.43309783936
dot_numba after compilation:1.20246815681
dot_numba_fast compilation:1.28146505356
dot_numba_fast after compilation:1.18621397018
dot_numba_parallel compilation:1.3401529789
dot_numba_parallel after compilation:1.44713497162
dot_numba_fast_parallel compilation:1.34533500671
dot_numba_fast_parallel after compilation:1.19588494301

依然没有看到加速的迹象.

#!/usr/bin/python
# -*- coding: utf-8 -*-
#####################################
# File name : test.py
# Create date : 2019-01-05 17:11
# Modified date : 2019-01-05 22:39
# Author : DARREN
# Describe : not set
# Email : [email protected]
#####################################
from __future__ import division
from __future__ import print_function

import numpy as np
from timeit import Timer
from numba import jit
from numba import njit

def dot(x,y):
    z = np.dot(x,y)
    return z

@jit
def dot_numba(x,y):
    z = np.dot(x,y)
    return z

@njit(fastmath=True)
def dot_numba_fast(x,y):
    z = np.dot(x,y)
    return z

@njit(parallel=True)
def dot_numba_parallel(x,y):
    z = np.dot(x,y)
    return z

@njit(fastmath=True, parallel=True)
def dot_numba_fast_parallel(x,y):
    z = np.dot(x,y)
    return z

if __name__ == '__main__':

    x = np.random.randn(4000,4000)
    y = np.random.randn(4000,4000)

    t1 = Timer('dot(x,y)','from __main__ import dot,x,y')
    t2 = Timer('dot_numba(x,y)','from __main__ import dot_numba,x,y')
    t3 = Timer('dot_numba(x,y)','from __main__ import dot_numba,x,y')
    t4 = Timer('dot_numba_fast(x,y)','from __main__ import dot_numba_fast,x,y')
    t5 = Timer('dot_numba_fast(x,y)','from __main__ import dot_numba_fast,x,y')

    t6 = Timer('dot_numba_parallel(x,y)','from __main__ import dot_numba_parallel,x,y')
    t7 = Timer('dot_numba_parallel(x,y)','from __main__ import dot_numba_parallel,x,y')

    t8 = Timer('dot_numba_fast_parallel(x,y)','from __main__ import dot_numba_fast_parallel,x,y')
    t9 = Timer('dot_numba_fast_parallel(x,y)','from __main__ import dot_numba_fast_parallel,x,y')

    take_time1 = t1.timeit(10)
    take_time2 = t2.timeit(10)
    take_time3 = t3.timeit(10)
    take_time4 = t4.timeit(10)
    take_time5 = t5.timeit(10)
    take_time6 = t6.timeit(10)
    take_time7 = t7.timeit(10)
    take_time8 = t8.timeit(10)
    take_time9 = t9.timeit(10)

    print("dot:%s" % take_time1)
    print("dot_numba compilation:%s" % take_time2)
    print("dot_numba after compilation:%s" % take_time3)
    print("dot_numba_fast compilation:%s" % take_time4)
    print("dot_numba_fast after compilation:%s" % take_time5)
    print("dot_numba_parallel compilation:%s" % take_time6)
    print("dot_numba_parallel after compilation:%s" % take_time7)
    print("dot_numba_fast_parallel compilation:%s" % take_time8)
    print("dot_numba_fast_parallel after compilation:%s" % take_time9)

output

dot:7.37525200844
dot_numba compilation:7.05400204659
dot_numba after compilation:7.02080202103
dot_numba_fast compilation:7.13517785072
dot_numba_fast after compilation:6.97198486328
dot_numba_parallel compilation:6.88464617729
dot_numba_parallel after compilation:6.97216892242
dot_numba_fast_parallel compilation:7.04679107666
dot_numba_fast_parallel after compilation:7.12123513222

numba 对于矩阵相乘似乎没有加速效果

猜你喜欢

转载自blog.csdn.net/DarrenXf/article/details/85863796
今日推荐
《美国对全球网络空间安全与发展的威胁和破坏》报告发布
火速冲上 GitHub 热榜 —— 开源编程语言、框架哪有这么可爱？
北京人形机器人创新中心发布全球首个纯电驱拟人奔跑的全尺寸人形机器人“天工”
LFOSSA 源来如此公开课 | 掌握云原生未来：CNCF 认证全面攻略与备考秘籍
周排行
循环神经网络（rnn）讲解
Tigao教程四：单独的关节运动
金蝶K3WISE15.0-注册套打教程
如何在Mac上配置Kubernetes
Android应用结束自身进程的方法
SpringMVC学习 十三 拦截器栈
中国驻洛杉矶总领馆举行新春招待会
HttpClient get post 发送
11 - three.js 笔记 - 绘制三维字体模型
Mysql递归获取某个父节点下面的所有子节点和子节点上的所有父节点
每日归档
更多
2024-05-01(4)
2024-04-30(1)
2024-04-29(40)
2024-04-28(0)
2024-04-27(56)
2024-04-26(39)
2024-04-25(22)
2024-04-24(36)
2024-04-23(26)
2024-04-22(39)

代码天地 © 2018 codetd.com

境外服务器   最新电视剧2022