cafew / emm_fbeta

Exceptional Model Mining in multi-label data

dssd-fbeta

Data section

Contains different datasets. Note that the data are in .arff format with a different style from the one used in http://mulan.sourceforge.net/. There are three types of attributes:

• numeric: for numerical attributes (the numerical pattern space will be the one presented in Kaytoue-2011.
• symbolic: for nominal attributes.
• boolean: for boolean attributes (special case of nominal attributes where only the presence of the item is considered in the pattern space). Note that in order to ignore attributes, the keyword ignore is used, as shown in data/reducedFlags.arff.

The boxplots of the label distributions of the six datasets in data are given in this sections. The dynamic beta is also plotted where the two parameters xbeta and lbeta are given in scripts/xlbetas.csv. Note that, for data/reducedFlags.arff the xbeta and lbeta parameters are computed automatically using the approach presented in the section 7.2 (page 17) of the paper.

Experiments section

xlbeta

The script given in xlbeta experiment show how to draw the boxplot with its dynamic beta plot for the desired dataset. Please read xp.sh to see how to plot the desired dataset. The used script (to customize manually xbeta and lbeta parameters) is given in script.py.

mcts

The main experiments results of the paper is given by the script xp.sh. Please comment the undesired files before launching the script.

Note that the last line of the script ./launch.sh ../../data/reducedFlags.arff numeric 1 contains a second parameter 1 which is used to test the MCTS method against the complete method via a boxplot as shown in the below image. Note that when testing against the complete method, the dataset should be small (the number of patterns is combinatorial).

For the exhaustive list of Figures please launch simply the script xp.sh. The list of results and Figures will be generated in experiments/mcts/results folder.

beam

This section give the results shown in the section 7.5 (page 19) of the paper using beam search algorithm. As in the beforehand section, in order to get the exhaustive list of figures, please launch xp.sh. Comment the undesired files before launching the script.