NumPy for Python 2.7 on 64-bit Windows

Question

I tried to get NumPy for Python 2.7 on Windows 64 bit, but the page http://www.lfd.uci.edu/~gohlke/pythonlibs/ that all references do not open on any of my devices.

Anywhere else can I find it?

Answer 1

I suggest WinPython , a Python 2.7 distribution for Windows with 32- and 64-bit versions.

This WinPython creator blog post explains why it is usually hard to find the 64-bit version of Windows NumPy:

According to experienced developers, there is no decent open (free) Fortran compiler for the Windows 64-bit platform. As a result, it is not possible to build NumPy or SciPy on this platform using only free, open source tools. That is why there are no official Windows 64 bit binaries for these two libraries. The only ready-to-use installers available there were prepared by Christoph Golke (using the Intel Fortran compiler, aka "ifort"), and these are clearly unofficial binaries. In addition, Christoph built two different installers for NumPy: one unoptimized and one optimized with the Intel Math Kernel Library (MKL), which provides better performance. And the Gohlke SciPy 64-bit package (the only one available online) requires the NumPy MKL. The problem is that, according to Christoph Golke, the MKL license does not allow me (or anyone else) to redistribute these binaries if I have not acquired such a license. It’s still not clear to me whether the end user will also require this license. I hope not. Let's pretend that. Besides, having carefully read the terms of the Intel MKL license, I am quite sure that I can redistribute the NumPy built on the basis of MKL, because it is simply a redistribution of runtime. Therefore, I think I will buy the Intel Fortran Compiler license (including MKL) in order to be able to rebuild NumPy and SciPy in the near future, but at the same time I will just redistribute the packages created by Christoph Golke.

Answer 2

I don’t know where else to find it, but I would like to mention that it’s actually not so difficult to build if you have a numpy source and an MSVC compiler.

Numpy builds using the python distutils package, and as such deals with manifest files. If you use VC9, you can probably go straight ahead.

I did this today using the VC10 compiler, and as such was supposed to modify the distutils package (msvc9compiler.py) so as not to process any manifest. I just commented on these lines. Then, before I built the package, I installed env var to point to my actual compiler:

set VS90COMNTOOLS=%VS100COMNTOOLS% c:\python27_64\python.exe setup.py build 

And after the build, I will find the numpy package in the build folder.

Numpy complains a lot about ATLAS and BLAS and more, but you end up with a compiled numpy that will run the tests successfully.