Source code for the ECML/PKDD 2014 Discovery Challenge. Here we present a simple ideia which was first place in two of the three tasks of the ECML/PKDD 2014 discovery challenge. See the website below for details on the challenge:

https://sites.google.com/site/predictivechallenge2014/home

• Flavio Figueiredo flaviov@dcc.ufmg.br

* Numpy >= 1.8.1 - http://scipy.org
* Pandas >= 0.13.1 - http://pandas.pydata.org
* Scikit-Learn >= 0.14.1 - http://scikit-learn.org
* Scipy >= 0.14.1 - http://scipy.org

As an optional package (for some of the ideias exploited in the paper but not used on the challenge), you will required pyksc:

* pyksc - http://github.com/flaviovdf/pyksc (head branch)

For faster training times we used the packages available with Anaconda (http://http://www.continuum.io). Anaconda make's use of MKL for multi-threaded matrix operations on the packages above. However, the code is pretty fast and should not take more than a few minutes even without MKL bindings.

Edit the config.py file with the approriate locations of the train/ and test/ folder (provided by the competition organizers).

After this, execute the myio.py script::

python myio.py

Execute the cross_val.py script to find the most suitable model. Use the --help option of the script for details on optional arguments.

Execute the predict.py script. It will perform cross-validation again (given the small runtime), but you may also supply manual parameters. Have a look at the options in using --help.