[PYTHON] Install scipy on Linux without internet connection

It is possible to copy the source code, but assuming that you cannot use installation tools that are premised on connecting to the Internet, install from the source.

$ python
Python 2.7.6 (default, May  1 2014, 11:34:02) 
[GCC 4.1.2 20080704 (Red Hat 4.1.2-46)] on linux2

As you can see, the version of python is 2.7.6.

numpy First, install from numpy.

numpy-1.8.1.tar.gz

Will be downloaded.

$ tar xvzf numpy-1.8.1.tar.gz
$ cd numpy-1.8.1/
$ python setup.py build --fcompiler=gnu95
# python setup.py install

It seems that gfortran was specified as the Fortran compiler. The last installation command was done with super-user privileges.

$ python
>>> import numpy

If there is no error, there is no problem.

Cython It seems that Cython is needed, so install it.

Cython-0.20.1.zip

Download and copy.

$ unzip Cython-0.20.1.zip
$ cd Cython-0.20.1/
# python setup.py install

Again, the last command was done with super-user privileges.

lapack It seems that lapack is also needed, so install it.

Edit some in the make.inc file.

FORTRAN= gfortran -m64 -fPIC

Here, -m64 is required to create 64-bit, and -fPIC is required to create a shared library.

$ make

Finally,

# cp liblapack.a libtmglib.a /usr/local/lib/

Install liblapack.a and libtmglib.a with a manual copy.

scipy Install with the following command.

# python setup.py install

If there are no errors as shown below, you can assume that the installation was successful.

>>> import scipy

ImportError: libblas.so: cannot open shared object file: No such file or directory

>>> import scipy.optimize

If you get the above error when you try, it is probably because libblas.so cannot be found in the library path.

$ export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/lib

Add the path to libblas.so to the value of LD_LIBRARY_PATH like this.

matplotlib Install as follows.

$ tar xvzf matplotlib-1.3.1.tar.gz
$ cd matplotlib-1.3.1/
$ cp ../distribute-0.6.28.tar.gz ./
$ python setup.py build
# python setup.py install

It seems to depend on distribute-0.6.28.tar.gz, so you have to put it in the matplotlib-1.3.1 / directory.

In addition, nose, dateutil, pyparsing, six, tornado, backports, etc. are required as dependencies. These can be done as long as the dependencies are resolved.