Fit interpretable models. Explain blackbox machine learning.
In the beginning machines learned in darkness, and data scientists struggled in the void to explain them.
Let there be light.
InterpretML is an open-source package that incorporates state-of-the-art machine learning interpretability techniques under one roof. With this package, you can train interpretable glassbox models and explain blackbox systems. InterpretML helps you understand your model’s global behavior, or understand the reasons behind individual predictions.
Interpretability is essential for:
Python 3.7+ | Linux, Mac, Windows
pip install interpret
# OR
conda install -c conda-forge interpret
EBM is an interpretable model developed at Microsoft Research*. It uses modern machine learning techniques like bagging, gradient boosting, and automatic interaction detection to breathe new life into traditional GAMs (Generalized Additive Models). This makes EBMs as accurate as state-of-the-art techniques like random forests and gradient boosted trees. However, unlike these blackbox models, EBMs produce exact explanations and are editable by domain experts.
Dataset/AUROC | Domain | Logistic Regression | Random Forest | XGBoost | Explainable Boosting Machine |
---|---|---|---|---|---|
Adult Income | Finance | .907±.003 | .903±.002 | .927±.001 | .928±.002 |
Heart Disease | Medical | .895±.030 | .890±.008 | .851±.018 | .898±.013 |
Breast Cancer | Medical | .995±.005 | .992±.009 | .992±.010 | .995±.006 |
Telecom Churn | Business | .849±.005 | .824±.004 | .828±.010 | .852±.006 |
Credit Fraud | Security | .979±.002 | .950±.007 | .981±.003 | .981±.003 |
Notebook for reproducing table
Interpretability Technique | Type |
---|---|
Explainable Boosting | glassbox model |
APLR | glassbox model |
Decision Tree | glassbox model |
Decision Rule List | glassbox model |
Linear/Logistic Regression | glassbox model |
SHAP Kernel Explainer | blackbox explainer |
LIME | blackbox explainer |
Morris Sensitivity Analysis | blackbox explainer |
Partial Dependence | blackbox explainer |
Let’s fit an Explainable Boosting Machine
from interpret.glassbox import ExplainableBoostingClassifier
ebm = ExplainableBoostingClassifier()
ebm.fit(X_train, y_train)
# or substitute with LogisticRegression, DecisionTreeClassifier, RuleListClassifier, ...
# EBM supports pandas dataframes, numpy arrays, and handles "string" data natively.
Understand the model
from interpret import show
ebm_global = ebm.explain_global()
show(ebm_global)
Understand individual predictions
ebm_local = ebm.explain_local(X_test, y_test)
show(ebm_local)
And if you have multiple model explanations, compare them
show([logistic_regression_global, decision_tree_global])
If you need to keep your data private, use Differentially Private EBMs (see DP-EBMs)
from interpret.privacy import DPExplainableBoostingClassifier, DPExplainableBoostingRegressor
dp_ebm = DPExplainableBoostingClassifier(epsilon=1, delta=1e-5) # Specify privacy parameters
dp_ebm.fit(X_train, y_train)
show(dp_ebm.explain_global()) # Identical function calls to standard EBMs
For more information, see the documentation.
EBMs include pairwise interactions by default. For 3-way interactions and higher see this notebook: https://interpret.ml/docs/python/examples/custom-interactions.html
Interpret EBMs can be fit on datasets with 100 million samples in several hours. For larger workloads consider using distributed EBMs on Azure SynapseML: classification EBMs and regression EBMs
InterpretML was originally created by (equal contributions): Samuel Jenkins, Harsha Nori, Paul Koch, and Rich Caruana
EBMs are fast derivative of GA2M, invented by: Yin Lou, Rich Caruana, Johannes Gehrke, and Giles Hooker
Many people have supported us along the way. Check out ACKNOWLEDGEMENTS.md!
We also build on top of many great packages. Please check them out!
plotly |
dash |
scikit-learn |
lime |
shap |
salib |
skope-rules |
treeinterpreter |
gevent |
joblib |
pytest |
jupyter
@article{nori2019interpretml, title={InterpretML: A Unified Framework for Machine Learning Interpretability}, author={Nori, Harsha and Jenkins, Samuel and Koch, Paul and Caruana, Rich}, journal={arXiv preprint arXiv:1909.09223}, year={2019} }Paper link
@inproceedings{caruana2015intelligible, title={Intelligible models for healthcare: Predicting pneumonia risk and hospital 30-day readmission}, author={Caruana, Rich and Lou, Yin and Gehrke, Johannes and Koch, Paul and Sturm, Marc and Elhadad, Noemie}, booktitle={Proceedings of the 21th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining}, pages={1721--1730}, year={2015}, organization={ACM} }Paper link
@inproceedings{lou2013accurate, title={Accurate intelligible models with pairwise interactions}, author={Lou, Yin and Caruana, Rich and Gehrke, Johannes and Hooker, Giles}, booktitle={Proceedings of the 19th ACM SIGKDD international conference on Knowledge discovery and data mining}, pages={623--631}, year={2013}, organization={ACM} }Paper link
@inproceedings{lou2012intelligible, title={Intelligible models for classification and regression}, author={Lou, Yin and Caruana, Rich and Gehrke, Johannes}, booktitle={Proceedings of the 18th ACM SIGKDD international conference on Knowledge discovery and data mining}, pages={150--158}, year={2012}, organization={ACM} }Paper link
@article{wang2022interpretability, title={Interpretability, Then What? Editing Machine Learning Models to Reflect Human Knowledge and Values}, author={Wang, Zijie J and Kale, Alex and Nori, Harsha and Stella, Peter and Nunnally, Mark E and Chau, Duen Horng and Vorvoreanu, Mihaela and Vaughan, Jennifer Wortman and Caruana, Rich}, journal={arXiv preprint arXiv:2206.15465}, year={2022} }Paper link
@inproceedings{zhang2019axiomatic, title={Axiomatic Interpretability for Multiclass Additive Models}, author={Zhang, Xuezhou and Tan, Sarah and Koch, Paul and Lou, Yin and Chajewska, Urszula and Caruana, Rich}, booktitle={Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery \& Data Mining}, pages={226--234}, year={2019}, organization={ACM} }Paper link
@inproceedings{tan2018distill, title={Distill-and-compare: auditing black-box models using transparent model distillation}, author={Tan, Sarah and Caruana, Rich and Hooker, Giles and Lou, Yin}, booktitle={Proceedings of the 2018 AAAI/ACM Conference on AI, Ethics, and Society}, pages={303--310}, year={2018}, organization={ACM} }Paper link
@article{lengerich2019purifying, title={Purifying Interaction Effects with the Functional ANOVA: An Efficient Algorithm for Recovering Identifiable Additive Models}, author={Lengerich, Benjamin and Tan, Sarah and Chang, Chun-Hao and Hooker, Giles and Caruana, Rich}, journal={arXiv preprint arXiv:1911.04974}, year={2019} }Paper link
@inproceedings{kaur2020interpreting, title={Interpreting Interpretability: Understanding Data Scientists' Use of Interpretability Tools for Machine Learning}, author={Kaur, Harmanpreet and Nori, Harsha and Jenkins, Samuel and Caruana, Rich and Wallach, Hanna and Wortman Vaughan, Jennifer}, booktitle={Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems}, pages={1--14}, year={2020} }Paper link
@article{chang2020interpretable, title={How Interpretable and Trustworthy are GAMs?}, author={Chang, Chun-Hao and Tan, Sarah and Lengerich, Ben and Goldenberg, Anna and Caruana, Rich}, journal={arXiv preprint arXiv:2006.06466}, year={2020} }Paper link
@inproceedings{pmlr-v139-nori21a, title = {Accuracy, Interpretability, and Differential Privacy via Explainable Boosting}, author = {Nori, Harsha and Caruana, Rich and Bu, Zhiqi and Shen, Judy Hanwen and Kulkarni, Janardhan}, booktitle = {Proceedings of the 38th International Conference on Machine Learning}, pages = {8227--8237}, year = {2021}, volume = {139}, series = {Proceedings of Machine Learning Research}, publisher = {PMLR} }Paper link
@inproceedings{ribeiro2016should, title={Why should i trust you?: Explaining the predictions of any classifier}, author={Ribeiro, Marco Tulio and Singh, Sameer and Guestrin, Carlos}, booktitle={Proceedings of the 22nd ACM SIGKDD international conference on knowledge discovery and data mining}, pages={1135--1144}, year={2016}, organization={ACM} }Paper link
@incollection{NIPS2017_7062, title = {A Unified Approach to Interpreting Model Predictions}, author = {Lundberg, Scott M and Lee, Su-In}, booktitle = {Advances in Neural Information Processing Systems 30}, editor = {I. Guyon and U. V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett}, pages = {4765--4774}, year = {2017}, publisher = {Curran Associates, Inc.}, url = {https://papers.nips.cc/paper/7062-a-unified-approach-to-interpreting-model-predictions.pdf} }Paper link
@article{lundberg2018consistent, title={Consistent individualized feature attribution for tree ensembles}, author={Lundberg, Scott M and Erion, Gabriel G and Lee, Su-In}, journal={arXiv preprint arXiv:1802.03888}, year={2018} }Paper link
@article{lundberg2018explainable, title={Explainable machine-learning predictions for the prevention of hypoxaemia during surgery}, author={Lundberg, Scott M and Nair, Bala and Vavilala, Monica S and Horibe, Mayumi and Eisses, Michael J and Adams, Trevor and Liston, David E and Low, Daniel King-Wai and Newman, Shu-Fang and Kim, Jerry and others}, journal={Nature Biomedical Engineering}, volume={2}, number={10}, pages={749}, year={2018}, publisher={Nature Publishing Group} }Paper link
@article{herman2017salib, title={SALib: An open-source Python library for Sensitivity Analysis.}, author={Herman, Jonathan D and Usher, Will}, journal={J. Open Source Software}, volume={2}, number={9}, pages={97}, year={2017} }Paper link
@article{morris1991factorial, title={}, author={Morris, Max D}, journal={Technometrics}, volume={33}, number={2}, pages={161--174}, year={1991}, publisher={Taylor \& Francis Group} }Paper link
@article{friedman2001greedy, title={Greedy function approximation: a gradient boosting machine}, author={Friedman, Jerome H}, journal={Annals of statistics}, pages={1189--1232}, year={2001}, publisher={JSTOR} }Paper link
@article{pedregosa2011scikit, title={Scikit-learn: Machine learning in Python}, author={Pedregosa, Fabian and Varoquaux, Ga{\"e}l and Gramfort, Alexandre and Michel, Vincent and Thirion, Bertrand and Grisel, Olivier and Blondel, Mathieu and Prettenhofer, Peter and Weiss, Ron and Dubourg, Vincent and others}, journal={Journal of machine learning research}, volume={12}, number={Oct}, pages={2825--2830}, year={2011} }Paper link
@online{plotly, author = {Plotly Technologies Inc.}, title = {Collaborative data science}, publisher = {Plotly Technologies Inc.}, address = {Montreal, QC}, year = {2015}, url = {https://plot.ly} }Link
@article{varoquaux2009joblib, title={Joblib: running python function as pipeline jobs}, author={Varoquaux, Ga{\"e}l and Grisel, O}, journal={packages. python. org/joblib}, year={2009} }Link
There are multiple ways to get in touch:
If a tree fell in your random forest, would anyone notice?