floriankrach/regretoptimalfederatedtransferlearning

Official implementation of "Regret-Optimal Federated Transfer Learning for Kernel Regression -- With Applications in American Option Pricing"

Regret Optimal Federated Transfer Learning

Official implementation of the paper Regret-Optimal Federated Transfer Learning for Kernel Regression with Applications in American Option Pricing.

Installation & Requirements

This code was executed using Python 3.8.

To install requirements, download this Repo and cd into it.

Then create a new environment and install all dependencies and this repo. With conda:

conda create --name FederatedTransferLearning python=3.8
conda activate FederatedTransferLearning
pip install -r requirements.txt

Usage

Use the run.py script to (parallely) run the experiments.

Example:

python code/run.py --params=params_list0 --NB_JOBS=1 --nb_runs=10 --DEBUG=1 --first_id=1 --get_overview=GTO_0

List of all flags:

• params: name of the params list (defined in config.py) to use for parallel run
• NB_JOBS: nb of parallel jobs to run with joblib
• nb_runs: nb of runs to do for each job
• first_id: First id of the given list / to start training of
• get_overview: name of the dict (defined in config.py) defining input for extras.get_training_overview
• conv_plot: name of the dict (defined in config.py) defining input for extras.plot_conv_analysis(_multiruns)
• SEND: whether to send results via telegram
• DEBUG: whether to run parallel in debug mode
• saved_models_path: path where the models are saved

run Heston / rough Heston experiments in paper:

python code/run.py --params=params_list6 --NB_JOBS=1 --nb_runs=100 --DEBUG=0 --first_id=1 --get_overview=GTO_6

python code/run.py --params=params_list7_1 --NB_JOBS=1 --nb_runs=100 --DEBUG=0 --first_id=1 --get_overview=GTO_7_1
python code/run.py --params=params_list7_2 --NB_JOBS=1 --nb_runs=100 --DEBUG=0 --first_id=1 --get_overview=GTO_7_2

run convergence experiments in paper:

python code/run.py --params=conv_params_list_2 --NB_JOBS=1 --nb_runs=10 --DEBUG=0 --first_id=1 --conv_plot=conv_plot_2
python code/run.py --conv_plot=conv_plot_2; python code/run.py --conv_plot=conv_plot_2_1; python code/run.py --conv_plot=conv_plot_2_2;  

additional convergence experiments:

python code/run.py --params=conv_params_list_3 --NB_JOBS=1 --nb_runs=10 --DEBUG=0 --first_id=1 --conv_plot=conv_plot_3
python code/run.py --params=conv_params_list_4 --NB_JOBS=1 --nb_runs=10 --DEBUG=0 --first_id=1 --conv_plot=conv_plot_4

---

License

This code can be used in accordance with the LICENSE.

---

Citation

If you use this code for your publications, please cite our paper: Regret-Optimal Federated Transfer Learning for Kernel Regression with Applications in American Option Pricing.

@misc{yang2023regretoptimal,
      title={Regret-Optimal Federated Transfer Learning for Kernel Regression with Applications in American Option Pricing}, 
      author={Xuwei Yang and Anastasis Kratsios and Florian Krach and Matheus Grasselli and Aurelien Lucchi},
      year={2023},
      eprint={2309.04557},
      archivePrefix={arXiv},
      primaryClass={cs.LG}
}

---