To run our experiments, execute
python3 experiments.py
in this folder. Note, that only a representative sample experiment is submitted. However, any expirement we did throughout the challenge can be expressed in a succinct manner using the versatile util.cross_validation function. It takes as input three arrays representing different models, kernels and training data respectively. util.cross_valdiation outputs a scores object that can be plotted using the util.plot_cross_val function. To classify test data use util.save_predictions.
For a usage sample see experiments.py. The repository contains cached kernel matrices for the Mismatch kernel with k=10 and k=12 (m=4 in both cases).
For the weighted degree kernel we implemented the computation of kernel function in C++. To compile the c++ code, such that they can be called form our pyhton code execute (in project root):
g++ -c -fPIC Code/c-functions/kernel-functions.cpp -Wextra -Wall -o Code/c-functions/fun.o && g++ -shared -Wl,-soname,Code/c-functions/kernel-functions.so -o Code/c-functions/kernel-functions.so Code/c-functions/fun.o