Skip to content

Cooperative Spatial Topologies for Autoencoder Training

License

Notifications You must be signed in to change notification settings

ALFA-group/lipizzaner-ae

Repository files navigation

Lipizzaner Autoencoder (Lipi-Ae)

Installation

Create virtual environment. E.g

python3 -m venv ~/.venvs/lipi_ae_gecco_24

Activate virtual environment. E.g

source ~/.venvs/lipi_ae_gecco_24/bin/activate

Install dependencies

pip install -r requirements.txt

Quick start

Binary Clustering

Create Binary Clustering problems

PYTHONPATH=src python src/aes_lipi/datasets/data_loader.py --n_dim 1000 --n_clusters 10

Test binary clustering problem and autoencoder

PYTHONPATH=src python src/aes_lipi/environments/binary_clustering.py --method=Autoencoder --dataset_name=binary_clustering_10_100_1000

Run Lipi-Ae on binary clustering problem

PYTHONPATH=src python src/aes_lipi/lipi_ae.py --configuration_file=tests/gecco_2024/configurations/binary_clustering/test_bc/binary_clustering_epoch_node_demo_lipi_ae.json

Experiment

Run experiments

 time PYTHONPATH=src python src/aes_lipi/utilities/gecco_experiments.py --configuration_directory tests/gecco_2024/configurations/binary_clustering/test_bc --sensitivity tests/gecco_2024/configurations/binary_clustering/test_bc/sensitivity_values.json

Update dataset in sensitivity_values.json key "dataset_name" by adding the new dataset to the list

Analyze data from --root_dir based on --param_dir parameters.

time PYTHONPATH=src python src/aes_lipi/utilities/analyse_data.py --root_dir out_binary_clustering --param_dir out_binary_clustering 
Compare ANN parameters

Save the parameters at every iteration Run Lipi-Ae with solution concept best_case

PYTHONPATH=src python src/aes_lipi/lipi_ae.py --dataset_name binary_clustering_10_100_1000  --environment AutoencoderBinaryClustering --epochs 3 --batch_size 400 --population_size 2 --ae_quality_measures L1 --solution_concept best_case --checkpoint_interval 1 --do_not_overwrite_checkpoint

Reference


@inproceedings{hemberg2024ae,
  title={Cooperative Spatial Topologies for Autoencoder Training},
  author={Hemberg, Erik and Toutouh, Jamal and O'Reilly, Una-May},
  booktitle={Proceedings of the Genetic and Evolutionary Computation Conference},
  year={2024}
}

About

Cooperative Spatial Topologies for Autoencoder Training

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages