Swift

A compiler for BLOG probabilistic programming language (http://bayesianlogic.github.io/).

Check the paper:

Proceeding of IJCAI 2016: http://www.ijcai.org/Proceedings/16/Papers/512.pdf

Arxiv: https://arxiv.org/abs/1606.09242

Code style:

1. double spaces for indentation
2. use linux line terminator
3. follow http://google-styleguide.googlecode.com/svn/trunk/cppguide.xml

Dependencies:

1. A C++ compiler (default GNU C++, VC++/Clang compatible)
2. The armadillo package: http://arma.sourceforge.net/
3. C++11 required
4. (recommended) BLAS and LAPACK

Armadillo Setup for OS X:

1. install armadillo library: brew install armadillo
2. link C++ libraries for OS X: xcode-select --install

Armadillo Setup for Ubuntu (Linux):

1. install required dependencies: apt-get install g++ cmake libopenblas-dev liblapack-dev
2. download armadillo for linux from its webpage: http://arma.sourceforge.net/download.html
3. unzip the tar.xz file, install armadillo according to section 5 and section 6 in README.txt

Armadillo Setup for Windows:

Download the VC++ solution config file from the armadillo webpage

How to Use:

1. Compile the Swift compiler: make compile (see compile.bat for windows)
2. Generate target C++ code:./swift -i <Model File> -o <Target C++ File> -e <Algorithm> (swift.exe in Windows). Run ./swift to see all the command line options
3. Compile your target code: g++ -o -O3 <target> <your C++ file> random/* -larmadillo (see swifty.sh).
4. Run your final binary!

Note if you are using visual studio under Windows, please compile inside visual studio to obtain the final binary (step 3).

Release and Pre-Compiled Binaries

Check our release version v0.1 and the binary files: https://github.com/lileicc/swift/releases/tag/v0.1

Documentation

Swift inherits all the syntax from BLOG with some extentions. Detailed documentations are under development. Please firstly check models under "example" directory to have a first impression. You can also check the documentations for BLOG for basic usage (http://bayesianlogic.github.io/pages/documentation.html).