How to fix Python Numpy/Pandas installation?

问题

I would like to install Python Pandas library (0.8.1) on Mac OS X 10.6.8. This library needs Numpy>=1.6.

I tried this

$ sudo easy_install pandas
Searching for pandas
Reading http://pypi.python.org/simple/pandas/
Reading http://pandas.pydata.org
Reading http://pandas.sourceforge.net
Best match: pandas 0.8.1
Downloading http://pypi.python.org/packages/source/p/pandas/pandas-0.8.1.zip#md5=d2c5c5bea971cd760b0ae6f6850fcb74
Processing pandas-0.8.1.zip
Running pandas-0.8.1/setup.py -q bdist_egg --dist-dir /tmp/easy_install-ckAMym/pandas-0.8.1/egg-dist-tmp-0mlL7t
error: Setup script exited with pandas requires NumPy >= 1.6 due to datetime64 dependency

So I tried to install Numpy

$ sudo easy_install numpy
Searching for numpy
Best match: numpy 1.6.2
Adding numpy 1.6.2 to easy-install.pth file

Using /Library/Python/2.6/site-packages
Processing dependencies for numpy
Finished processing dependencies for numpy

So I tried again

$ sudo easy_install pandas

But the problem is still the same !

error: Setup script exited with pandas requires NumPy >= 1.6 due to datetime64 dependency

I run Python

$ python
Python 2.6.1 (r261:67515, Jun 24 2010, 21:47:49) 
[GCC 4.2.1 (Apple Inc. build 5646)] on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>> import numpy as np
>>> np.__version__
'1.2.1'

So Numpy 1.6 doesn't seems to be installed correctly !

I tried to install Numpy 1.6 with pip (instead of easy_install)...

$ sudo pip install numpy
Requirement already satisfied (use --upgrade to upgrade): numpy in /Library/Python/2.6/site-packages
Cleaning up...

I added --upgrade flag

$ sudo pip install numpy --upgrade
Requirement already up-to-date: numpy in /Library/Python/2.6/site-packages
Cleaning up...

$ sudo pip install pandas
Downloading/unpacking pandas
  Downloading pandas-0.8.1.zip (1.9MB): 1.9MB downloaded
  Running setup.py egg_info for package pandas
    pandas requires NumPy >= 1.6 due to datetime64 dependency
    Complete output from command python setup.py egg_info:
    pandas requires NumPy >= 1.6 due to datetime64 dependency

----------------------------------------
Command python setup.py egg_info failed with error code 1 in /tmp/pip-build/pandas
Storing complete log in /Users/MyUsername/Library/Logs/pip.log

I also tried to install binary version of Numpy http://sourceforge.net/projects/numpy/files/ numpy-1.6.2-py2.6-python.org-macosx10.3.dmg but it fails !!! (installer said me that numpy 1.6.2 can't be install on this disk. Numpy requires python.org Python 2.6 to install.

回答1:

Don't know if you solved the problem but if anyone has this problem in future.

$python
>>import numpy
>>print(numpy)

Go to the location printed and delete the numpy installation found there. You can then use pip or easy_install

回答2:

I had this exact problem.

The issue is that there is an old version of numpy in the default mac install, and that pip install pandas sees that one first and fails -- not going on to see that there is a newer version that pip herself has installed.

If you're on a default mac install, and you've done pip install numpy --upgrade to be sure you're up to date, but pip install pandas still fails due to an old numpy, try the following:

$ cd /System/Library/Frameworks/Python.framework/Versions/2.7/Extras/lib/python/
$ sudo rm -r numpy
$ pip install pandas

This should now install / build pandas.

To check it out what we've done, do the following: start python, and import numpy and import pandas. With any luck, numpy.__version__ will be 1.6.2 (or greater), and pandas.__version__ will be 0.9.1 (or greater).

If you'd like to see where pip has put (found!) them, just print(numpy) and print(pandas).

回答3:

I work with the guys that created Anaconda Python. You can install multiple versions of python and numpy without corrupting your system python. It's free and open source (OSX, linux, Windows). The paid packages are enhancements on top of the free version. Pandas is included.

conda create --name np17py27 anaconda=1.4 numpy=1.7 python=2.7
export PATH=~/anaconda/envs/np17py27/bin:$PATH

If you want numpy 1.6:

conda create --name np16py27 anaconda=1.4 numpy=1.6 python=2.7

Setting your PATH sets up where to find python and ipython. The environments (np17py27) can be named whatever you would like.

回答4:

This worked for me under 10.7.5 with EPD_free-7.3-2 from Enthought:

Install EPD free, then follow the step in the following link to create .bash_profile file.

http://redfinsolutions.com/blog/creating-bashprofile-your-mac

And add the following to the file.

PATH="/Library/Frameworks/Python.framework/Versions/Current/bin:$(PATH)}"
export PATH

Execute the following command in Terminal

$ sudo easy_install pandas

When finished, launch PyLab and type:

In [1]: import pandas

In [2]: plot(arange(10))

This should open a plot with a diagonal straight line.

回答5:

If you're like me and you don't like the idea of deleting things that were part of the standard system installation (which others have suggested) then you might like the solution I ended up using:

1. Get Homebrew - it's a one-line shell script to install!
2. Edit your .profile, or whatever is appropriate, and put /usr/local/bin at the start of your PATH so that Homebrew binaries are found before system binaries
3. brew install python - this installs a newer version of python in /usr/local
4. pip install pandas

This worked for me in OS X 10.8.2, and I can't see any reason it shouldn't work in 10.6.8.

回答6:

You probably have another Numpy version installed on your system, try to query your numpy version and retrieve it if your distribution does not support it.
aka debian/unbuntu/Mint version can query mostly from dpkg package manger : dpkg --get-selections | egrep -i "numpy", you can see actual Numpy version.

• Some having apt can either asking to removing it by doing this: apt-get remove numpy.

• Some having distribution like Fedora, RedHat and any compatible release under RedHat model can use rpm as well to query the installation.

• This is happening by telling to Numpy installer to install itself in current /usr/local/lib/python[VERSION]/dist-packages over Linux env and c:[...]\python[VERSION]\site-packages for windows. Having probably One version of Numpy installed in /usr/local/python[VERSION]/dist-packages, this one will be instantiated first.

• .pth file hold information about path location of specific python module, but erasing a component from packages may corrupt it...

Be careful, and you will have to remove the package and all it's dependency... really painful in some case.

Visiting lunchad.net may save you time sometimes they had new versions from some packages.

回答7:

I had the same problem and, in my case, the problem was that python was looking for packages in some ordered locations, first of all the default computer one where default old packages are.

To check what your python is looking for you can do:

>>> import sys
>>> print '\n'.join(sys.path)

This was outputting the directory '/System/Library/Frameworks/Python.framework/Versions/2.7/Extras/lib/python' before pip or brew or port folders.

The simple solution is:

export PYTHONPATH="/Library/Python/2.7/site-packages:$PYTHONPATH"

This worked well for me, I advise you to add this line to your home bash_profile file for the next time. Remember that sys.path is built using the current working directory, followed by the directories in the PYTHONPATH environment variable. Then there are the installation-dependent default dirs.

回答8:

If you are using a version of enthought python (EPD) you might want to go directly to your site-packages and reinstall numpy. Then try to install pandas with pip. You will have to modify your installation prefix for that.

If the problem persists (as it did with me) try downloading pandas tar ball, unpack it in your site packages and run setup.py install from your pandas directory.

If you got your dependencies right you can import pandas and check it imports smoothly.

回答9:

he easiest way to install Pandas, like almost every other package for Python, is with pip.

Many packages (including Pandas) require a compiler, and a bunch of third-party DLLs, and many Windows users don't know how to deal with that. That's exactly why the "wheel" format was created: so packages can upload pre-built binaries.

Not every project has pre-built binary wheels for Windows yet. But you can look at Christoph Gohlke's site and find wheels for all of the most popular ones. Just follow the instructions on that page to download the wheel file and install it with pip.

But in the case of Pandas, you don't have to do that. They have wheels on their download page, and uploaded to PyPI. And the documentation tells you to use these. (Well, it first suggests you use Anaconda/Miniconda, but if you want a stock Python, use pip and the packages on PyPI.) it worked for me ...on windows 7 64 bit ,python 3.4

来源：https://stackoverflow.com/questions/12436979/how-to-fix-python-numpy-pandas-installation