Fractal-based decomposition experiments using Zellij

Experiments for the paper "A fractal-based decomposition framework for continuous optimization".

Zellij

Experiments use the in development Zellij version. Please intall the Zellij version from develop_t branch.

$ pip install https://github.com/ThomasFirmin/zellij/archive/develop_t.zip

Benchmarks data

Shift vectors, rotation matrices and shuffled dimensions are stored in cec2020/input_data from the CEC2020 github, and socco2011/input_data. Functions are defined in cec2020/functions.py and socco2011/functions.py, and imported from Zellij benchmarks.py.

The wk_price_SP500.txt dataset, can be downloaded from original dataset. (Man-Fai Leung: Datasets for Portfolio Optimization. Mendeley (2022). https://doi.org/10.17632/G5579MMC9K.2)

Algorithms

Algorithms are defined in algorithms.py using Zellij.

Algorithms are instantited within test.py and test_portfolio.py:

$ algo = algorithm(decision_variables, loss_function)

Default parameters are used.

Experiments

Files test.py and test_portfolio contains code launching experiments.* Use following commands in a terminal to launch an experiment :

$ python3 ./test.py -a FDA -b cec2020 -f 0 -d 10

• The option -a, --algorithm is for selecting an algorithm within FDA, DIRECT, DIRECTL, DIRECTR, SOO, DIRECTBs, FDABs, FDADBs, SOOBs, FDAC, FDAD. If no option is passed, then experiments are computed for all algorithms.
• The option -b, --benchmark is for selecting a benchmark within cec2020, socco2011.
• The option -f, --function is for selecting a function ID within the selected benchmark. Within [0,9] for cec2020, and [0,10] for socco2011. If no option is passed, then experiments are computed for all functions.
• The option -d, --dimension is for selecting a dimension within 10, 15, 20, 30, 50, 100. If no option is passed, then experiments are computed for all dimmensions.

For SP500 portfolio optimization: wk_price_SP500.txt must be downloaded and included in the working directory.

$ python3 ./test_portfolio.py -a FDA

• The option -a, --algorithm is for selecting an algorithm within FDA, DIRECT, DIRECTL, DIRECTR, SOO, DIRECTBs, FDABs, FDADBs, SOOBs, FDAC, FDAD. If no option is passed, then experiments are computed for all algorithms.

All experimental data will be stored in individual folders for each algorithm, function and dimension, within experiments_cec2020, experiments_socco2011 and experiments_sp500. The total size can reach 30 Gb.

Analysis

Analysis are made in analysis.ipynb.

Experimental datasets

Firmin, Thomas, 2023, "Experimental results for the paper "A fractal-based decomposition framework for continuous optimization"", https://doi.org/10.57745/0JEUEK, Recherche Data Gouv, V1, UNF:6:lW7nJ0QKwy3UZSdtxZVz/w== [fileUNF]