Repository contains reimplementation of code and experiments, based on this repository, which implements following paper. The purpouse of this repo is to reproduce results and simplify further research using python3 as tool.

We manage packages via anaconda. The easiest setup would be:

• install conda
• clone repository
• enter repository
• conda env create -f environment.yml
• conda activate myenv

Run: jupyter notebook

• Experiments and visualizations for different data types are kept in experiments folder.
• .py files contain preprocesing tools and models

Some models are already precomputed and added to repo with best hyperparameters found using gridsearch. In such case models will be loaded and can be used for further needs.

• To recalculate experiment clear .dill files under paths experiments/dataset/precomputed/cv, experiments/dataset/precomputed/grid.
• Apply modifications to Model_Comparison.ipynb
• Re-evaluate notebook