From 2be8342e56f03771c0523445cd7c69857be7617b Mon Sep 17 00:00:00 2001 From: Pat Alt <55311242+pat-alt@users.noreply.github.com> Date: Fri, 19 May 2023 08:49:11 +0200 Subject: [PATCH] README done --- README.md | 68 +++++++++++++++++++++++++++++++++++++++++++++++++++++- README.qmd | 54 ------------------------------------------- 2 files changed, 67 insertions(+), 55 deletions(-) delete mode 100644 README.qmd diff --git a/README.md b/README.md index dff821d3..dd43df3b 100644 --- a/README.md +++ b/README.md @@ -1,3 +1,69 @@ # ECCCo -[](https://github.com/pat-alt/ECCCo.jl/actions/workflows/CI.yml?query=branch%3Amain) + + +*Energy-Constrained Counterfactual Explanations.* + +This work is currently undergoing peer review. This README is therefore only meant to provide reviewers access to the code base. The code base will be made public after the review process. + +## Inspecting the Package Code + +This code base is structured as a Julia package. The package code is located in the `src/` folder. + +## Inspecting the Code for Experiments + +We used [Quarto](https://quarto.org/) notebooks for prototyping and running experiments. The notebooks are located in the `notebooks/` folder, separated by dataset: + +- [Linearly Separable](notebooks/linearly_separable.qmd) +- [Moons](notebooks/moons.qmd) +- [Circles](notebooks/circles.qmd) +- [MNIST](notebooks/mnist.qmd) +- [GMSC](notebooks/gmsc.qmd) + +Instead of looking at the notebooks directly, you may choose to browse the HTML book contained inside the `docs` folder. The book is automatically generated from the notebooks and includes all code chunks and their outputs. It is currently not possible to view the book online, but you can download the `docs/` folder and open the `index.html` file in your browser. + +## Inspecting the Results + +All results have been carefully reported either in the paper itself or in the supplementary material. In addition, we have released our results as binary files. These will be made publicly available after the review process. + +## Reproducing the Results + +To reproduce the results, you need to install the package, which will automatically install all dependencies. Since the package is not publicly registered and you are looking at an anonymous repository that [cannot be cloned](https://anonymous.4open.science/faq#download), unfortunately, it is not possible to easily install the package and reproduce the results at this stage of the review process. + +However, provided that the package is indeed installed, you can reproduce the results by either running the experiments in the `experiments/` folder or using the notebooks listed above for a more interactive process. + +### Command Line + +The `experiments/` folder contains separate Julia scripts for each dataset and a [run_experiments.jl](experiments/run_experiments.jl) that calls the individual scripts. You can either cun these scripts inside a Julia session or just use the command line to execute them as described in the following. + +To run the experiment for a single dataset, (e.g. `linearly_separable`) simply run the following command: + +```shell +DATANAME=linearly_separable +julia experiments/run_experiments.jl +``` + +We use the following identifiers: + +- `linearly_separable` (*Linearly Separable* data) +- `moons` (*Moons* data) +- `circles` (*Circles* data) +- `mnist` (*MNIST* data) +- `gmsc` (*GMSC* data) + +To run all experiments at once you can instead just specify `DATANAME=all`. + +Pre-trained versions of all of our black-box models have been archived as `Pkg` [artifacts](https://pkgdocs.julialang.org/v1/artifacts/) and are used by default. Should you wish to retrain the models as well, simply run the following command: + +```shell +DATANAME=linearly_separable +RETRAIN=true +julia experiments/run_experiments.jl +``` + +When running the experiments from the command line, the parameter choices used in the main paper are applied by default. To have control over these choices, we recommend you instead rely on the notebooks. + +### Notebooks + +To run the notebooks and ensure that all package dependencies are installed, you need to clone this repo and open it on your device. The first cell in each notebook sets up the environment. You may have to [instantiate](https://pkgdocs.julialang.org/v1/api/#Pkg.instantiate) the local environment once. Should you prefer working with Jupyter notebooks instead of Quarto, you can easily [convert](https://quarto.org/docs/tools/vscode-notebook.html#converting-notebooks) them through a single command. + diff --git a/README.qmd b/README.qmd deleted file mode 100644 index 371fab96..00000000 --- a/README.qmd +++ /dev/null @@ -1,54 +0,0 @@ ---- -format: - commonmark: - variant: -raw_html+tex_math_dollars - wrap: none - mermaid-format: png -crossref: - fig-prefix: Figure - tbl-prefix: Table -bibliography: https://raw.githubusercontent.com/pat-alt/bib/main/bib.bib -output: asis -jupyter: julia-1.8 -execute: - freeze: auto - eval: true - echo: true - output: false ---- - -# ECCCo - -*Counterfactual Explanations and Algorithmic Recourse in Julia.* - -This work is currently undergoing peer review. This README is therefore only meant to provide reviewers access to the code base. The code base will be made public after the review process. - -## Inspecting the Package Code - -This code base is structured as a Julia package. The package code is located in the `src` folder. - -## Inspecting the Code for Experiments - -We used [Quarto] notebooks for prototyping and running experiments. The notebooks are located in the `notebooks` folder, separated by dataset: - -- [Linearly Separable](notebooks/linearly_separable.qmd) -- [Moons](notebooks/moons.qmd) -- [Circles](notebooks/circles.qmd) -- [MNIST](notebooks/mnist.qmd) -- [GMSC](notebooks/gmsc.qmd) - -Instead of looking at the notebooks directly, you may choose to browse the HTML book contained inside the `docs` folder. The book is automatically generated from the notebooks and includes all code chunks and their outputs. It is currently not possible to view the book online, but you can download the `docs` folder and open the `index.html` file in your browser. - -## Reproducing the Results - -To reproduce the results, you need to install the package, which will automatically install all dependencies. Since the package is not publicly registered and you are looking at an anonymous repository that [cannot be cloned](https://anonymous.4open.science/faq#download), unfortunately, it is not possible to easily install the package and reproduce the results at this stage of the review process. - -However, provided that the package is indeed installed, you can reproduce the results by running the experiments in the `experiments` folder. The experiments are structured as Julia scripts. You can run the experiments by executing the following command: - - - - - - - - -- GitLab