BNFinder2 - new version available!

We have just released a new version of BNFinder. Since original release in 2009, it contains a number of bugfixes and a whole new set of functions. Most notable additions are:

• Support for classification with Bayesian networs including cross-validation, ROC curve plotting and classification of new examples
• Inclusion of the Mutual Information Test (MIT) scoring function for static/dynamic BNs with support for continuous variables
• Support for parallel computation on multiple CPU machines
• Possibility to pre-define the desired false-positive error-rate during training

The new version is available for download from Python Package Index and from our launchpad project page.

You can also download documentation in pdf format as a single file or in separate parts:

• User's manual
• Tutorial
• Supplementary methods

Documentation sources and the exemplary input files are available in the source package in the doc/ folder.

BNFinder - usage statistics

As a part of our efforts on releasing the new BNfinder version we have looked at some statistics of BNfinder usage. Our input data was simply the logs of our web interface (we cannot gather all download data from different sites such as pypi and launchpad...). All in all, within the three years sinc BNFinder got published, our server served a bit more than 750 requests from a bit over 220 users. We were curious where are our users from, so we did a little analysis on the top-level domain parts of the registered e-mail addresses. More than half (125) users have an e-mail address somewhere on gmail.com or yahoo.com which prevents identification but for the remaining users we mapped their location based on their TLD (mapping .edu to US). The results are interesting:

BNfinder seems to be used across almost-all continents (although it clearly needs some publicity in South America and Africa ;)).

BNFinder - Bayesian Network Finder

BNF toolbox allows for Bayesian Network reconstruction. It can be used for both dynamic and static networks. It is written in python, and distributed under GNU GPL Library version 2.

The stable version of the software itself can be downloaded from here. A more recent, development version can be found on launchpad. It requires python version > 2.4, (python 2.5 is recommended). Python 3.0 is not supported yet. Permanent installation is described in the installation istruction, you can also run it without installing, just unpack the archive, go to the created directory and type:

python bnf --help

Documentation:

• Tutorial on using BNFinder in pdf file for printing, html format or pdf in zip archive with data examples
• Description of methods in html or pdf
• Users manual in html or pdf
• Whole documentation in a pdf version

If you do not want to install the software, try our Web interface. The interface is very simple (see screenshot). In order to test it, you only need to provide an input file containing observed experimental data (example file) and your e-mail address. The resulting network (in a SIF format readable by Cytoscape) will be sent to your e-mail address.

If you have any questions or comments, please use our mailinglist, or contact us directly at: [ dojer | bartek ]@mimuw.edu.pl,