Code of FedRain and Frog for VLDB 2022

For the paper with secure CG protocol, please take a look at the technical report

Run cd FedRain && poetry install to install the python dependencies.

Goto FedRain/scripts, run call_slave.py in one docker container and start_slave.py in another container. Make sure to:

1. change the slave address in the call_slave.py accordingly.
2. Set enabled=True for the LogFile class in both call_slave.pyand start_slave.py.
3. Start a postgres server in docker and set the corresponding address when initializing the LogFile class in call_slave.pyand start_slave.py.

Then run call_slave.py and the log will be populated to the database. Run FedRain/analysis.ipynb to compute the time cost from the log.

Similar to Table 3, but make sure to disable the log by setting enabled=False for the LogFile class in both call_slave.pyand start_slave.py. This is due to logging will take a significant amount of time. The end to end running time will be printed out once call_slave.py finished.

Run RecallK.ipynb to reproduce the data for Figure 4.

Run cd Frog && pip install -r requirements.txt to install the python dependencies.

Goto Experiments/processors/diabetes, run train_debug_master.py in one docker container and train_debug_slave.py in another container. The computation and encryption time would be printed out.

Run Experiments/com_time.py to compute the time cost for communication.

For Figure 3, we duplicate the data points and features (e.g. x = np.tile(x,(50, 1))) in Experiments/processors/diabetes/__init__.py to get the scalability performance of Frog.

Run Experiments/adhoc/LinearComb_Diabetes.ipynb and Experiments/adhoc/LinearComb_BreastCancer.ipynb to reproduce the data for Figure 4. The Recall@k plots are summarized in Experiments/adhoc/LC_Corruption_Plot.ipynb.

Run Experiments/adhoc/LinearComb_Adults_DisparateImpact.ipynb.