The repo consists of implementation of SUDS for two different algorithms, D3 and OCDD. The repository is organized as follows
- D3 directory contains all code and scripts related to D3
- OCDD directory contains all code and scripts related to OCDD
D3.py and OCDD.py are the original algorithm implementations of their respective authors. Our modifications contain "_SUDS" suffix to the original algorithms. Each algorithm is executed twenty-five times using a script called run_all_D3.sh for D3 and run_all_OCDD.sh for OCDD. We also execute the same algorithm without the interleaved test then train method.
A gather data script takes the output of the bash script and creates a CSV file for the accuracy, amount of training samples required and the number of drifts for each dataset.
The plots.ipynb file contains the plots created for the paper. The data is obtained from the xlsx files within each folder.
- To reproduce the results simply run the shell script run_all_OCDD.sh or run_all_D3.sh and all the output files are saved
- The gather_data.py gathers the results and writes the output to the corresponding .csv file