SIRUS.jl

SIRUS.jl: Interpretable Machine Learning via Rule Extraction - Published in JOSS (2023)

https://github.com/rikhuijzer/sirus.jl

Science Score: 98.0%

This score indicates how likely this project is to be science-related based on various indicators:

✓
CITATION.cff file
Found CITATION.cff file
✓
codemeta.json file
Found codemeta.json file
✓
.zenodo.json file
Found .zenodo.json file
✓
DOI references
Found 5 DOI reference(s) in README and JOSS metadata
✓
Academic publication links
Links to: joss.theoj.org
○
Committers with academic emails
○
Institutional organization owner
✓
JOSS paper metadata
Published in Journal of Open Source Software

Keywords

explainable-ai explainable-ml machine-learning

Keywords from Contributors

standardization meshing pde interpretability parallel simulations correlation graphics fluxes hydrology

Last synced: 6 months ago · JSON representation ·

Repository

Interpretable Machine Learning via Rule Extraction

Basic Info

Host: GitHub
Owner: rikhuijzer
License: mit
Language: Julia
Default Branch: main
Homepage: https://sirus.jl.huijzer.xyz/
Size: 1.2 MB

Statistics

Stars: 38
Watchers: 2
Forks: 3
Open Issues: 19
Releases: 12

Topics

explainable-ai explainable-ml machine-learning

Created over 3 years ago · Last pushed 7 months ago

Metadata Files

Readme License Citation

README.md

Visual representation of the algorithm which converts decision trees to rule sets. Created with DALL·E 3 and Photopea

SIRUS.jl

This package is a pure Julia implementation of the Stable and Interpretable RUle Sets (SIRUS) algorithm. The algorithm was originally created by Clément Bénard, Gérard Biau, Sébastien Da Veiga, and Erwan Scornet (Bénard et al., 2021). SIRUS.jl has implemented both classification and regression, but we found that performance is generally best on classification tasks.

The main benefit of this algorithm is that it is fully explainable. This differs from model-agnostic explainability techniques such as SHAP, which convert the model to a simplified representation. However, the complex model is still used for predictions, which can lead to hidden biases or reliability issues. The SIRUS algorithm fixes this by using a simplified model for both for prediction and explanation.

Installation

```julia julia> ]

pkg> add SIRUS ```

Getting Started

This package defines two rule-based models that satisfy the Machine Learning Julia MLJ.jl interface. The models are StableRulesClassifier and StableRulesRegressor:

Example

```julia julia> using MLJ, SIRUS

julia> X, y = make_blobs(200, 10; centers=2);

julia> X Tables.MatrixTable{Matrix{Float64}} with 200 rows, 10 columns, and schema: :x1 Float64 :x2 Float64 :x3 Float64 :x4 Float64 :x5 Float64 :x6 Float64 :x7 Float64 :x8 Float64 :x9 Float64 :x10 Float64

julia> y 200-element CategoricalArrays.CategoricalArray{Int64,1,UInt32}: 2 1 1 ⋮ 2 1 2

julia> model = StableRulesClassifier();

julia> mach = machine(model, X, y);

julia> fit!(mach);

julia> mach.fitresult StableRules model with 7 rules: if X[i, :x5] < -1.552594 then 0.129 else 0.0 + if X[i, :x8] < 0.72402614 then 0.117 else 0.0 + if X[i, :x2] < 7.1123967 then 0.123 else 0.0 + if X[i, :x8] < 8.840833 then 0.115 else 0.0 + if X[i, :x9] < 7.985747 then 0.0 else 0.001 + if X[i, :x7] < 6.4651833 then 0.107 else 0.0 + if X[i, :x7] < 2.2220817 then 0.119 else 0.024 and 2 classes: [1, 2]. Note: showing only the probability for class 2 since class 1 has probability 1 - p. ```

This is a basic example, in most cases you want to tune the max_depth, max_rules, and lambda hyperparameters. See ?StableRulesClassifier, ?StableRulesRegressor, or the API documentation for more information about the models and their hyperparameters. A full guide through binary classification can be found in the Simple Binary Classification example.

Citation

bibtex @article{huijzer2023sirus, title={{SIRUS.jl}: Interpretable Machine Learning via Rule Extraction}, author={Huijzer, Rik and Blaauw, Frank and den Hartigh, Ruud JR}, journal={Journal of Open Source Software}, volume={8}, number={90}, pages={5786}, year={2023}, doi={10.21105/joss.05786} }

Documentation

Documentation is at sirus.jl.huijzer.xyz.

Contributing

Thank you for your interest in contributing to SIRUS.jl! There are multiple ways to contribute.

Questions and Bug Reports

For questions or bug reports, you can open an issue. Questions can also be asked at the Julia forum or by sending a mail to github@huijzer.xyz. Tag @rikh in the forum to ensure a quick reply.

Pull Requests

To submit patches, use pull requests (PRs) here on GitHub. In general:

Try to keep PRs limited to one feature or bug; otherwise they become hard to review/verify.
Try to use the code style that is used in the rest of the codebase. See also the Code Style Blue.
Try to update documentation when updating code, but feel free to leave documentation updates for a separate PR.
When possible, make PRs as easily reversible as possible. Any change that would be easily reversible later provides little risk and can, therefore, more easily be merged.

As long as the PR moves the codebase forward, merging will likely happen.

Owner

Name: Rik Huijzer
Login: rikhuijzer
Kind: user

Website: https://huijzer.xyz
Repositories: 196
Profile: https://github.com/rikhuijzer

JOSS Publication

SIRUS.jl: Interpretable Machine Learning via Rule Extraction

Published

October 12, 2023

DOI

10.21105/joss.05786

Volume 8, Issue 90, Page 5786

Authors

Rik Huijzer

University of Groningen, Groningen, the Netherlands

Frank Blaauw

Researchable, Assen, the Netherlands

Ruud J.R. den Hartigh

University of Groningen, Groningen, the Netherlands

Editor

Mehmet Hakan Satman

View PDF Review Thread Software Archive

Citation (CITATION.cff)

cff-version: "1.2.0"
authors:
- family-names: Huijzer
  given-names: Rik
  orcid: "https://orcid.org/0000-0001-9445-8466"
- family-names: Blaauw
  given-names: Frank
  orcid: "https://orcid.org/0000-0002-6588-5079"
- family-names: Hartigh
  given-names: Ruud J. R.
  name-particle: den
  orcid: "https://orcid.org/0000-0002-0094-8307"
contact:
- family-names: Huijzer
  given-names: Rik
  orcid: "https://orcid.org/0000-0001-9445-8466"
doi: 10.5281/zenodo.8398350
message: "If you use this software, please cite our article in the Journal of Open Source Software."
preferred-citation:
  authors:
  - family-names: Huijzer
    given-names: Rik
    orcid: "https://orcid.org/0000-0001-9445-8466"
  - family-names: Blaauw
    given-names: Frank
    orcid: "https://orcid.org/0000-0002-6588-5079"
  - family-names: Hartigh
    given-names: Ruud J. R.
    name-particle: den
    orcid: "https://orcid.org/0000-0002-0094-8307"
  date-published: 2023-10-12
  doi: 10.21105/joss.05786
  issn: 2475-9066
  issue: 90
  journal: Journal of Open Source Software
  publisher:
    name: Open Journals
  start: 5786
  title: "SIRUS.jl: Interpretable Machine Learning via Rule Extraction"
  type: article
  url: "https://joss.theoj.org/papers/10.21105/joss.05786"
  volume: 8
title: "SIRUS.jl: Interpretable Machine Learning via Rule Extraction"

GitHub Events

Total

Create event: 3
Commit comment event: 2
Issues event: 1
Watch event: 7
Delete event: 1
Push event: 10
Pull request event: 5

Last Year

Create event: 3
Commit comment event: 2
Issues event: 1
Watch event: 7
Delete event: 1
Push event: 10
Pull request event: 5

Committers

Last synced: 7 months ago

All Time

Total Commits: 251
Total Committers: 5
Avg Commits per committer: 50.2
Development Distribution Score (DDS): 0.052

Past Year

Commits: 7
Committers: 1
Avg Commits per committer: 7.0
Development Distribution Score (DDS): 0.0

Top Committers

Name	Email	Commits
Rik Huijzer	r**r@p**e	238
dependabot[bot]	4****]	6
github-actions[bot]	4****]	5
Okon Samuel	3****l	1
Jose Storopoli	j**e@s**o	1

Committer Domains (Top 20 + Academic)

storopoli.io: 1 pm.me: 1

Issues and Pull Requests

Last synced: 6 months ago

All Time

Total issues: 40
Total pull requests: 54
Average time to close issues: about 2 months
Average time to close pull requests: 6 days
Total issue authors: 8
Total pull request authors: 6
Average comments per issue: 1.73
Average comments per pull request: 0.61
Merged pull requests: 48
Bot issues: 0
Bot pull requests: 14

Past Year

Issues: 0
Pull requests: 2
Average time to close issues: N/A
Average time to close pull requests: about 1 hour
Issue authors: 0
Pull request authors: 2
Average comments per issue: 0
Average comments per pull request: 0.5
Merged pull requests: 1
Bot issues: 0
Bot pull requests: 1

View more stats

Top Authors

Issue Authors

rikhuijzer (22)
gdalle (6)
ablaom (5)
sylvaticus (2)
OkonSamuel (1)
ericphanson (1)
JuliaTagBot (1)
Zapiano (1)

Pull Request Authors

rikhuijzer (36)
dependabot[bot] (16)
github-actions[bot] (5)
jbytecode (2)
storopoli (2)
OkonSamuel (1)

Top Labels

Issue Labels

enhancement (3)

Pull Request Labels

dependencies (16)

Packages

Total packages: 1
Total downloads:
- julia 9 total

Total dependent packages: 1
Total dependent repositories: 0
Total versions: 13

juliahub.com: SIRUS

Interpretable Machine Learning via Rule Extraction

Homepage: https://sirus.jl.huijzer.xyz/
Documentation: https://docs.juliahub.com/General/SIRUS/stable/
License: MIT
Latest release: 2.0.1
published about 2 years ago

Versions: 13
Dependent Packages: 1
Dependent Repositories: 0
Downloads: 9 Total

Rankings

Dependent repos count: 9.9%

Average: 36.9%

Dependent packages count: 38.9%

Stargazers count: 45.1%

Forks count: 53.5%

Last synced: 6 months ago

Dependencies

.github/workflows/CI.yml actions

actions/checkout v2 composite
julia-actions/cache v1 composite
julia-actions/julia-buildpkg latest composite
julia-actions/julia-runtest v1 composite
julia-actions/setup-julia v1 composite

.github/workflows/CompatHelper.yml actions

julia-actions/setup-julia v1 composite

.github/workflows/Docs.yml actions

actions/checkout v2 composite
julia-actions/julia-buildpkg v1 composite
julia-actions/julia-docdeploy v1 composite

.github/workflows/TagBot.yml actions

JuliaRegistries/TagBot v1 composite