Installation

If you get errors at any point of this installation instructions get please in touch with me so I can fix it!

Dependencies

ATLAS

For Linux users, the easiest way to obtain ATLAS is via a package manager. For ubuntu, the package needed is called libatlas-base-dev. Mac OS X users already have the necessary libraries installed. If you want to do the installation yourself, there is installation information on the web.

Scons

The build process for arac works via scons. Most package managers will have a scons package. You can also download the source from the website or use setuptools easy_install. For pip install the sc0ns package. Make sure you have at least version 1.2.0 installed. You can check via:

$ scons --version
SCons by Steven Knight et al.:
	script: v1.2.0.d20090113.r3897, 2009/01/13 06:45:54, by scons on scons-dev
	engine: v1.2.0.d20090113.r3897, 2009/01/13 06:45:54, by scons on scons-dev
Copyright (c) 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009 The SCons Foundation

Python with development headers

You can get Python python.org – make sure to install the development headers as well (e.g. python-devel on ubuntu). 2.5+ is needed.

Swig

You can get swig here. However, it is preinstalled on many Unix systems. Also, probably every package manager out there has it. arac has successfully been built with swig 1.3.36. You can check your version with

$ swig -version

SWIG Version 1.3.36
[...]

Google Testing Framework

Attention You need to use version 1.4, which is not straightforward to find on the website.

Since arac is developed test driven, it is best to also run those tests. For this, you will have to compile the google testing framework from source. Download the latest version from their download page and unpack it. In the source directory run the following:

$ ./configure
$ make
$ make check
$ sudo make install

You will be asked for your superuser password in the last step.

Getting arac

You can either download arac (e.g. here) directly form github, clone it or fork it and clone your fork. e.g. you can do:

$ git clone git://github.com/bayerj/arac.git

Building arac

Change into the source directory and run

$ scons

Note: If you are seeing ‘ImportError: /usr/lib/liblapack.so.3gf: undefined symbol: ATL_chemv’ error, run
sudo apt-get remove libopenblas-base . See http://stackoverflow.com/questions/8917977/installing-lapack-for-numpy for details.

After that, you should find either a file libarac.so (Linux) or libarac.dylib in that directory. Make sure to build a symbolic link to it from one of the directories on your library path. For most users, this will do:

 arac/ $ sudo ln -s libarac.so /usr/local/lib/libarac.so

or this:

 arac/ $ sudo ln -s `pwd`/libarac.so /usr/local/lib/

(Ubuntu and Debian users might need to reload the list of system-wide library paths. However, there is still a good chance that the symbolic link won’t work. Copying the library directly to /usr/local/lib/ overcomes this problem.).

For Mac OS X, use libarac.dylib. After that, run the tests in the source directory:

 arac/ $ test-arac 

You should get a whole list of tests, that either pass or fail. Right now, the LSTMLayer is known to fail (but it’s okay).

Accessing arac from python

To make sure that the python bindings are working, make sure that the arac package located at arac/src/python/arac is under the python path. If you create a symbolic link or update the PYTHONPATH environment variable is up to you. E.g.:

 arac/ $ export PYTHONPATH=$PYTHONPATH:`pwd`/src/python

After that try:

$ python
>>> import arac

If this works, you are set.

Using arac with PyBrain

arac is mainly being used in conjunction with PyBrain. If you have been coming from the PyBrain docs to this place, you are now finished, too – PyBrain does not care where exactly arac resides.