Bagging for LID estimation

This repository has been developed for the paper titled "On the Use of Bagging for Local Intrinsic Dimensionality Estimation" with preprint published on arXiv, as well as part of a master's thesis project viewable at Master's Thesis.

• To recreate the results, plots, and figures present in the publication, first follow the Installation steps. Then do either of the following:

  • Follow the instructions at Reproducibility to resample new datasets and perform the LID estimation experiments in the publication from scratch.
  • Follow the instructions at Reproducibility to load ready-made light experiment objects from light experiments, then recreate the plots and figures in the preferred style. Note that these are made only for plotting purposes and have no longer the necessary data attached to recreate the experiments from scratch.

• To use the Bagging_for_LID package for your own LID estimation experiments, or to examine experiment objects, first install the package via the Installation steps. Then follow the instructions at Tutorial for the package.

Installation

Install the Bagging_for_LID package and its requirements.

To begin, clone the main repository to a selected folder

git clone https://github.com/Campello-Lab/Bagging_for_LID.git

Navigate to selected folder

cd Bagging_for_LID

Install requirements. Python version $\geq 3.11$ required.

pip install -r requirements.txt

Install package

pip install -e .

Reproducibility

Recreating results and figures from the publication On the Use of Bagging for Local Intrinsic Dimensionality Estimation

• The recreate_results_notebook Jupyter notebook file contains detailed, step-by-step instructions on how to recreate the results and figures present in the paper from scratch, or by loading already computed experiment objects.
• The recreate_example_figures Jupyter notebook file can be used to recreate the plots in the Introduction section of the publication.
• The recreate_results Python file can be used to recreate the same results without the use of a Jupyter notebook.

Tutorial for the package

Using the Bagging_for_LID package to run your own LID estimation experiments

• The single_experiment jupyter notebook file can be used to learn how to use the repository for examining or performing single LID estimation experiments, or multiple ones at once for a range of parameter combinations.

Contact

For questions or support, please open an issue or contact: krp@imada.sdu.dk.