arokem / python-matlab-bridge

A simple Python => MATLAB(R) interface and a matlab_magic for ipython
BSD 2-Clause "Simplified" License
335 stars 96 forks source link

Python-MATLAB(R) Bridge and IPython matlab magic

Build Status Coverage Status Latest Version License Join the chat at https://gitter.im/arokem/python-matlab-bridge

A python interface to call out to Matlab(R). Original implementation by Max Jaderberg. His original repo of the project can be found here, but please note that the development of the two repositories has significantly diverged.

This implementation also includes an IPython matlab_magic extension, which provides a simple interface for weaving python and Matlab code together (requires ipython > 0.13).

Installation

pymatbridge can be installed from PyPI:

$ pip install pymatbridge

If you intend to use the Matlab magic extension, you'll also need IPython.

Finally, if you want to handle sparse arrays, you will need to install Scipy. This can also be installed from PyPI, or using distributions such as Anaconda or Enthought Canopy

Usage

To use the pymatbridge you need to connect your python interpreter to a Matlab session. This is done in the following manner:

from pymatbridge import Matlab
mlab = Matlab()

This creates a matlab session class instance, into which you will be able to inject code and variables, and query for results. By default, when you use start, this will open whatever gets called when you type matlab in your Terminal, but you can also specify the location of your Matlab application when initializing your matlab session class:

mlab = Matlab(executable='/Applications/MATLAB_R2011a.app/bin/matlab')

You can then start the Matlab server, which will kick off your matlab session, and create the connection between your Python interpreter and this session:

mlab.start()

which will return True once connected.

results = mlab.run_code('a=1')

Should now run that line of code and return a results dict into your Python namespace. The results dict contains the following fields:

{u'content': {u'code': u'a=1',
 u'datadir': u'/private/tmp/MatlabData/',
 u'figures': [],
 u'stdout': u'\na =\n\n     1\n\n'},
 u'success': u'true'}

In this case, the variable a is available on the Python side, by using the get_variable method:

mlab.get_variable('a')

You can run any MATLAB functions contained within a .m file of the same name. For example, to call the function jk in jk.m:

%% MATLAB
function lol = jk(args)
    arg1 = args.arg1;
    arg2 = args.arg2;
    lol = arg1 + arg2;
end

you would call:

res = mlab.run_func('path/to/jk.m', {'arg1': 3, 'arg2': 5})
print(res['result'])

This would print 8.

You can shut down the MATLAB server by calling:

mlab.stop()

Tip: you can execute MATLAB code at the beginning of each of your matlab sessions by adding code to the ~/startup.m file.

Octave support & caveats

A pymatbridge.Octave class is provided with exactly the same interface as pymatbridge.Matlab:

from pymatbridge import Octave
octave = Octave()

Rather than looking for matlab at the shell, this will look for octave. As with pymatbridge.Matlab, you can override this by specifying the executable keyword argument.

Rather than ~/startup.m, Octave looks for an ~/.octaverc file for commands to execute before every session. (This is a good place to manipulate the runtime path, for example).

Requires Version 3.8 or higher. Notice: Neither the MXE 3.8.1 nor the Cygwin 3.8.2 version is compatible on Windows. No Windows support will be available until a working version of Octave 3.8+ with Java support is released.

Matlab magic:

The Matlab magic allows you to use pymatbridge in the context of the IPython notebook format.

%load_ext pymatbridge

These lines will automatically start the matlab session for you. Then, you can simply decorate a line/cell with the '%matlab' or '%%matlab' decorator and write matlab code:

%%matlab
a = linspace(0.01,6*pi,100);
plot(sin(a))
grid on
hold on
plot(cos(a),'r')

If %load_ext pymatbridge doesn't work for you use:

import pymatbridge as pymat
pymat.load_ipython_extension(get_ipython(), matlab='/your_matlab_installation_dir/bin/matlab')

More examples are provided in the examples directory

Building the pymatbridge messenger from source

The installation of pymatbridge includes a binary of a mex function to communicate between Python and Matlab using the 0MQ messaging library. This should work without any need for compilation on most computers. However, in some cases, you might want to build the pymatbridge messenger from source. To do so, you will need to follow the instructions below:

Install zmq library

Please refer to the official guide on how to build and install zmq. On Ubuntu, it is as simple as sudo apt-get install libzmq3-dev. On Windows, suggest using the following method:

After zmq is installed, make sure you can find the location where libzmq is installed. The library extension name and default location on different systems are listed below.

Platform library name Default locations
MacOS libzmq.dylib /usr/lib or /usr/local/lib
Linux libzmq.so.3 /usr/lib or /usr/local/lib
Windows libzmq.dll C:\Program Files\ZeroMQ 3.2.4\bin

If you specified a prefix when installing zmq, the library file should be located at the same prefix location.

The pymatbridge MEX extension needs to be able to locate the zmq library. If it's in a standard location, you may not need to do anything; if not, there are two ways to accomplish this:

Using the dynamic loader path

One option is to set an environment variable which will point the loader to the right directory.

On MacOS, you can do this by adding the following line to your .bash_profile (or similar file for your shell):

export DYLD_LIBRARY_PATH=$DYLD_LIBRARY_PATH:<Path to your zmq lib directory>

On Linux, add the following line to your .bash_profile (or similar file for your shell):

export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:<Path to your zmq lib directory>

On Windows, add the install location of libzmq.dll to the PATH environment variable. On Windows 7+, typing "environment variables" into the start menu will bring up the apporpriate Control Panel links.

Pointing the binary at the right place

Another option is to modify the MEX binary to point to the right location. This is preferable in that it doesn't change loader behavior for other libraries than just the pymatbridge messenger.

On MacOS, you can do this from the root of the pymatbridge code with:

install_name_tool -change /usr/local/lib/libzmq.3.dylib <Path to your zmq lib directory>/libzmq.3.dylib messenger/maci64/messenger.mexmaci64

On Linux, you can add it to the RPATH:

    patchelf --set-rpath <Path to your zmq lib directory> messenger/mexa64/messenger.mexa64

Install pyzmq

After step 1 is finished, please grab the latest version of pyzmq and follow the instructions on the official page. Note that pymatbridge is developed with pyzmq 14.0.0 and older versions might not be supported. If you have an old version of pyzmq, please update it.

Install pymatbridge

After the steps above are done, you can install pymatbridge. Download the zip file of the latest release. Unzip it somewhere on your machine.

For Matlab:

cd messenger
# edit local.cfg in the directory for your platform
python make.py matlab
cd ..
python setup.py install

For Octave:

cd messenger/octave
# edit local_octave.cfg in the directory for your platform
python make.py octave
cd ..
python setup.py

This should make the python-matlab-bridge import-able.

Warnings

Python communicates with Matlab via an ad-hoc zmq messenger. This is inherently insecure, as the Matlab instance may be directed to perform arbitrary system calls. There is no sandboxing of any kind. Use this code at your own risk.