NumPy 是一个基础科学的计算包。很多的科学计算特别是向量处理的时候会用到。 因为用到了,而且安装的过程中出现了一些问题,所以记录下来,便于备查。
平台: windows 8.1 64位
Python: 2.7.9 32位(已包含pip命令)
无Microsoft的VS环境
打开powershell,(事先将Python以及Python的Script的目录导入环境变量PATH中),输入
pip install numpy
会出现:
Unable to find vcvarsall.bat
的错误消息。
原因是:the Python package being installed contains the source code for a native extension module (.pyd), but does not have a pre-built copy of the module. The Python packages highlighted at pythonwheels.com have already been updated by their developers to include pre-built binaries, but many are still only available as source code.(来自微软官网)
我自己的理解是:要编译的代码中用到了C的代码,而这个包的代码发布的时候没有预先编译成Python的二进制文件,所以安装过程中需要使用C complier 编译这些模块。
所以,解决的方法就是安装 Microsoft Visual C++ Compiler for Python 2.7 。
安装成功之后,继续执行
pip install numpy
出现了下列的错误:
RuntimeError: Broken toolchain: cannot link a simple C program
具体原因是为什么,我也没看懂,不知道为什么,但是找到了解决这个问题的方法:
在%PythonHome%Lib\distutils下,(默认是C:\Python27\Lib\distutils)修改msvc9compiler.py文件:
# embed the manifest # XXX - this is somewhat fragile - if mt.exe fails, distutils # will still consider the DLL up-to-date, but it will not have a # manifest. Maybe we should link to a temp file? OTOH, that # implies a build environment error that shouldn‘t go undetected. # change by YQ to install numpy in 2015/1/16 mfinfo = None #<===# mfinfo = self.manifest_get_embed_info(target_desc, ld_args)
找到这些代码,应该在648行左右,将箭头右方的代码更改成箭头左方的代码。
然后再运行:
pip install numpy
终于OK了。
附测试程序:
# import the necessary packages import numpy as np def chi2_distance(histA, histB, eps = 1e-10): # compute the chi-squared distance d = 0.5 * np.sum([((a - b) ** 2) / (a + b + eps) for (a, b) in zip(histA, histB)]) Sum = 0; for (a,b) in zip(histA, histB): E = ((a - b) ** 2) / (a + b + eps) Sum += E print Sum/2 # return the chi-squared distance return d d = chi2_distance([0,1,2,3,4],[4,3,2,1,0]) # print d