DT-GO

This repository provides code for DT-GO (Delay Tolerant Gossiped Optimization).

We provide code for the main experiments in the papers:

Decentralized Optimization in Networks with Arbitrary Delays

Decentralized Optimization in Time-Varying Networks with Arbitrary Delays

Running the code

Please see requirements.txt to install the appropriate python requirements. Scienceplots is only required to make publication-quality plots, but is not needed to run the experiments.

To reproduce the experiments in Decentralized Optimization in Networks with Arbitrary Delays, navigate to the experiments folder and run python p_experiment.py and python lambda_experiment.py respectively.

• The code in p_experiment.py will run in less than a minute in any modern laptop.
• The code in lambda_experiment.py may take about 30 minutes to run in any modern laptop.

The code will produce the results for the paper figures for both experiments as-is. To produce the figures, navigate to the experiments folder and run python plot_p_experiment.py and python plot_lambda_experiment.py respectively.

To reproduce the experiments in Decentralized Optimization in Time-Varying Networks with Arbitrary Delays, navigate to the experiments folder and run python experiment_regression.py with the desired configurations.

Alternatively, run sh runall.sh to run all the experiments in parallel. This may take several days to complete depending on your hardware. It takes about 2 days to run all experiments in a AMD Ryzen Threadripper PRO 5955WX 16-Cores machine with 128 GB of RAM.

To produce the figures, navigate to the experiments folder and run python plot_experiment_p_err.py, for example.

References

If you use the code, please consider citing the following papers:

@misc{ortega2024decentralized_b,
      title={Decentralized Optimization in Time-Varying Networks with Arbitrary Delays}, 
      author={Tomas Ortega and Hamid Jafarkhani},
      year={2024},
      eprint={2405.19513},
      archivePrefix={arXiv},
      primaryClass={cs.LG}
}

@misc{ortega2024decentralized,
      title={Decentralized Optimization in Networks with Arbitrary Delays}, 
      author={Tomas Ortega and Hamid Jafarkhani},
      year={2024},
      eprint={2401.11344},
      archivePrefix={arXiv},
      primaryClass={math.OC}
}