Avalanche Boilerplate

This is a boilerplate for creating a new continual learning project using the Avalanche library. It includes a basic structure for the project, as well as some useful tools and libraries.

Getting Started

To use the code, first clone the repository:

git clone https://github.com/Atenrev/avalanche_boilerplate.git
cd avalanche_boilerplate

Then, assuming you have Python 3.9 set up, install the required libraries:

pip install -r requirements.txt

Running the Code

To run the continual learning experiments, use the following command:

main.py [-h]

For example, to run a continual learning experiment on Split MNIST with a simple MLP model using lfw, use the following command:

python main.py --strategy lfw --dataset split_mnist --model simple_mlp --output_dir results/ --alpha=1.0 --lr 0.001 --batch_size 128 --epochs 4 --seeds 42 69 1714

This will run the experiment with seeds 42, 69 and 1714, and save the results to results/dataset_name/{strategy}_on_{model}_{custom_parameters}_lr({args.lr})_bs({args.batch_size})_epochs({args.epochs})/. In this case, the results will be saved to results/split_mnist/lfw_on_simple_mlp_alpha(1.0)"_lr(0.001)_bs(128)_epochs(4)/. Inside this folder, you will find a folder for each seed, and inside each of these folders you will find the logs in CSV format inside the logs folder.

Visualizing the results

You can generate plots of the results using the generate_report.py script. For example, to generate a report for the results of the previous example, use the following command:

python generate_report.py --experiments_path results/split_mnist/lfw_on_simple_mlp_alpha(1.0)_lr(0.001)_bs(128)_epochs(4) --metrics Accuracy_On_Trained_Experiences --plot_individual_metrics --create_table --create_comparison_plots

This command will generate plots and tables for the specified metrics and save them in the corresponding experiment folder. The --experiments_path argument specifies the path to the experiment results, and the --metrics argument specifies the metrics to plot and compare. The --plot_individual_metrics flag will plot the metrics for each experiment individually, the --create_table flag will create a table with the final results, and the --create_comparison_plots flag will create comparison plots.

License

This project is licensed under the MIT License - see the LICENSE file for details.

References

This repository is based on the Avalanche library. If you use this code in your research, please consider citing the following paper:

@article{carta2023avalanche,
  title={Avalanche: A pytorch library for deep continual learning},
  author={Carta, Antonio and Pellegrini, Lorenzo and Cossu, Andrea and Hemati, Hamed and Lomonaco, Vincenzo},
  journal={Journal of Machine Learning Research},
  volume={24},
  number={363},
  pages={1--6},
  year={2023}
}