Classifier Adjacency Visualisation

Demo script for visualising behaviour, (dis)similarity or complementarity of binary classifiers in response to a common dataset.

The recipe itself is simple:

• compute rank correlation of scores (Kendall's tau)
• visualise with multidimensional scaling (MDS plot)

Please reference our paper

Kinnunen, T., Nautsch, A., Sahidullah, M., Evans, N., Wang, X., Todisco, M., Delgado, H., Yamagishi J, Lee, K.A.,
"Visualizing Classifier Adjacency Relations: A Case Study in Speaker Verification and Voice Anti-Spoofing", 
in Proc. Interspeech, 2021.

License: https://creativecommons.org/licenses/by/4.0/

Installation (Linux)

After installing conda:

conda create -n classifier-adjacency python=3.7
conda activate classifier-adjacency
pip install sklearn matplotlib pandas h5py
git clone https://gitlab.eurecom.fr/nautsch/pybosaris pybosaris-source
ln -s pybosaris-source/pybosaris pybosaris

Use

python classifier_adjacency_demo.py

This creates two tikz/pgfplot files:

asv.tikz
asv_legend.tikz

Tweak it a bit: figure 2(a) of our paper results