Before getting started, make sure to install the dpsched package under ./simulator.
./microbench_single_block.py contains the script to reproduce single static block workload experiments with various DP scheduling policies, for both epsilon-delta DP and Renyi DP.
./microbench_multi_block.py is the dynamic multiple block counterpart of microbench_single_block.py/
One can run them with:
cd ./evaluation/microbenchmark
pypy ./microbench_single_block.py
pypy ./microbench_multi_block.py(caveat: the multiple block script may take several hours to finish).
The experiment results are saved under ./evaluation/microbenchmark/exp_results/some_work_space_name
microbenchmark_figures_single_block.ipynb and microbenchmark_figures_multiple_block.ipynb show how to reproduce microbenchmark figures. You can read a notebook directly via nbviewer without executing it. Note that these two notebooks download existing microbenchmark results and plot them directly. Therefore, we don't need to rerun microbenchmark experiments.