Installing SciPy and NumPy using pip

问题:

I'm trying to create required libraries in a package I'm distributing. It requires both the SciPy and NumPy libraries. While developing, I installed both using

apt-get install scipy

which installed SciPy 0.9.0 and NumPy 1.5.1, and it worked fine.

I would like to do the same using pip install - in order to be able to specify dependencies in a setup.py of my own package.

The problem is, when I try:

pip install 'numpy==1.5.1'

it works fine.

But then

pip install 'scipy==0.9.0'

fails miserably, with

raise self.notfounderror(self.notfounderror.__doc__)

numpy.distutils.system_info.BlasNotFoundError:

Blas (http://www.netlib.org/blas/) libraries not found.

Directories to search for the libraries can be specified in the

numpy/distutils/site.cfg file (section [blas]) or by setting

the BLAS environment variable.

How do I get it to work?

回答1:

I am assuming Linux experience in my answer; I found that there are three prerequisites to getting pip install scipy to proceed nicely.

Go here: Installing SciPY

Follow the instructions to download, build and export the env variable for BLAS and then LAPACK. Be careful to not just blindly cut'n'paste the shell commands - there will be a few lines you need to select depending on your architecture, etc., and you'll need to fix/add the correct directories that it incorrectly assumes as well.

The third thing you may need is to yum install numpy-f2py or the equivalent.

Oh, yes and lastly, you may need to yum install gcc-gfortran as the libraries above are Fortran source.

回答2:

This worked for me on Ubuntu 14.04:

sudo apt-get install libblas-dev liblapack-dev libatlas-base-dev gfortran
pip install scipy

回答3:

you need the libblas and liblapack dev packages if you are using Ubuntu.

aptitude install libblas-dev liblapack-dev
pip install scipy

回答4:

Since the previous instructions for installing with yum are broken here are the updated instructions for installing on something like fedora. I've tested this on "Amazon Linux AMI 2016.03"

sudo yum install atlas-devel lapack-devel blas-devel libgfortran
pip install scipy

回答5:

I was working on a project that depended on numpy and scipy. In a clean installation of Fedora 23, using a python virtual environment for Python 3.4 (also worked for Python 2.7), and with the following in my setup.py (in the setup() method)

setup_requires=[
    'numpy',
],
install_requires=[
    'numpy',
    'scipy',
],

I found I had to run the following to get pip install -e . to work:

pip install --upgrade pip

and

sudo dnf install atlas-devel gcc-{c++,gfortran} subversion redhat-rpm-config

The redhat-rpm-config is for scipy's use of redhat-hardened-cc1 as opposed to the regular cc1

回答6:

What operating system is this? The answer might depend on the OS involved. However, it looks like you need to find this BLAS library and install it. It doesn't seem to be in PIP (you'll have to do it by hand thus), but if you install it, it ought let you progress your SciPy install.

回答7:

On windows, using python 3.5, I managed to install scipy by using conda not pip:

conda install scipy

回答8:

in my case, upgrading pip did the trick. Also, I've installed scipy with -U parameter (upgrade all packages to the last available version)