muggin/string-kernels

Implementation of the "Text Classification using String Kernels" publication by Lodhi et al. (KTH DD2434 Project)

String Kernels

Implementation of the "Text Classification using String Kernels" publication by Lodhi et al. Code was written mainly in Python with some parts moved to Cython for performance gains. The final report can be found here.

This project was carried out as part of the DD2434 "Advanced Machine Learning" course at KTH Royal Institute of Technology.

Contributors

• F. Franzen (github: flammi)
• B. Godefroy (github: BGodefroyFR)
• W. Kryściński (github: muggin)
• V. Polianskii (github: vlpolyansky)

Data

Files in the data directory:

• train_data and test_data - original Reuters dataset split (Modified Apte) and (Pickled)
• train_data_clean and test_data_clean - preprocessed and cleaned dataset (Pickled)
• train_data_small and test_data_small - trimmed dataset prepared for experiments (Pickled)
• precomp_kernels - directory with precomputed SSK gram matrices
• approx - directory with precomputed approximated-SSK files

Setup

Before using SSK kernel compile Cython code using:

python setup.py build_ext --inplace