bandyt

This README describes a collection of python scripts for the Bayes Network (BN) structure learning problem collectively called BNOmics. The software allows for a purely data driven BN structure learning and is best used in combination with a python shell, such as Ipython, for on-the-fly experimentation and research. It could, however, be easily used for stand alone scripts such as the included example ( example.py ).

Due to the inherent unpredictability of the demands imposed by the various -Omics (as in genomics or metabolomics, etc.) projects using BNOmics, it is meant to serve as a prototypical research platform, modified upon necessity to address the HPC needs, rather than a narrow purpose, typical desktop application.

Requirements: Python interpreter - required. Numpy module - required. Graphviz - optional but recomended for graphical rendering. GCC - optional but needed for performance boost. make - optional, needed for compilation.

Installation: No particular installation procedure is necessary. However to enable faster compute routines compilation of cpp source in the folder containing this project is necessary. The folder contains Makefile that will tell the compiler what to do as long as 'make' utility and g++ compiler are present. Having navigated to Bnomics folder in the terminal run

touch ofext.cpp
make

This procedure should update C++ extension to the current architechture and make optimized routines available.

Example script: To start with the example please open a terminal and navigate to the BNOmics folder. Once there, you can call example.py as a standard python script with a filename argument:

python example.py african_americans.csv

The two example data files african_americans.csv and european_americans.csv are provided in the collection.

After the execution of example.py completes, the structure of reconstructed BN will be saved in dotfile.dot and can be rendered with graphviz as follows:

dot -Tpdf dotfile.dot -o outpdf.pdf

If Graphviz is properly installed the rendering procedure will be called automatically generating outpdf.pdf upon execution of example.py , and the above manual invocation of the rendering procedure will not be neccessary.

Now you can open outpdf.pdf with any pdf viewer for investigation of the results.

Feel free to open example.py with your editor of choice and view the contents. This file contains the most typical and simple example of a workflow for data driven BN reconstruction. This little script can be easily modified and tuned using the comments provided in the file.

Interactive use: In an interactive environment you can usually examine the contents of the included files as follows:

import bnomics
help(bnomics)
help(bnomics.dutils)
help(bnomics.bnutils)

and so on.

The typical workflow will be identical to the example.py script with the additional benefit of further details and information available for examination. For example, a BN can be modified by hand, its structure can be viewed as a list, a different search method can be applied or even constructed, etc.

//=============================================================
//(c) 2011 Distributed under MIT-style license. 
//(see LICENSE.txt or visit http://opensource.org/licenses/MIT)
//=============================================================