November 13, 2019

643 words 4 mins read

interpretml/interpret

interpretml/interpret

Fit interpretable machine learning models. Explain blackbox machine learning.

repo name interpretml/interpret
repo link https://github.com/interpretml/interpret
homepage
language C++
size (curr.) 3049 kB
stars (curr.) 2399
created 2019-05-03
license MIT License

InterpretML - Alpha Release

License Python Version Package Version Build Status Coverage Maintenance

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 python package for training interpretable machine learning models and explaining blackbox systems. Interpretability is essential for:

  • Model debugging - Why did my model make this mistake?
  • Detecting bias - Does my model discriminate?
  • Human-AI cooperation - How can I understand and trust the model’s decisions?
  • Regulatory compliance - Does my model satisfy legal requirements?
  • High-risk applications - Healthcare, finance, judicial, …

Historically, the most interpretable machine learning models were not very accurate, and the most accurate models were not very interpretable. Microsoft Research has developed an algorithm called the Explainable Boosting Machine (EBM)* which has both high accuracy and interpretability. EBM uses modern machine learning techniques like bagging and boosting to breathe new life into traditional GAMs (Generalized Additive Models). This makes them as accurate as random forests and gradient boosted trees, and also enhances their intelligibility and editability.

Notebook for reproducing table

Dataset/AUROC Domain Logistic Regression Random Forest XGBoost Explainable Boosting Machine
Adult Income Finance .907±.003 .903±.002 .922±.002 .928±.002
Heart Disease Medical .895±.030 .890±.008 .870±.014 .916±.010
Breast Cancer Medical .995±.005 .992±.009 .995±.006 .995±.006
Telecom Churn Business .804±.015 .824±.002 .850±.006 .851±.005
Credit Fraud Security .979±.002 .950±.007 .981±.003 .975±.005

In addition to EBM, InterpretML also supports methods like LIME, SHAP, linear models, partial dependence, decision trees and rule lists. The package makes it easy to compare and contrast models to find the best one for your needs.

* EBM is a fast implementation of GA2M. Details on the algorithm can be found here.


Installation

Python 3.5+ | Linux, Mac OS X, Windows

pip install -U interpret

Getting Started

Let’s fit an Explainable Boosting Machine

from interpret.glassbox import ExplainableBoostingClassifier

ebm = ExplainableBoostingClassifier()
ebm.fit(X_train, y_train)

# 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)

Global Explanation Image

Understand individual predictions

ebm_local = ebm.explain_local(X_test, y_test)
show(ebm_local)

Local Explanation Image

And if you have multiple models, compare them

show([logistic_regression, decision_tree])

Dashboard Image

Example Notebooks

Roadmap

Currently we’re working on:

  • R language interface (R is currently a WIP. Basic EBM classification can be done via the ebm_classify & ebm_predict_proba functions, but the predictions are a bit less accurate than in python. No plotting included yet, but other R plotting tools can do a basic job visualizing EBM models)
  • Missing Values Support
  • Improved Categorical Encoding
  • Interaction effect purification (see citations for details)

…and lots more! Get in touch to find out more.

Contributing

If you are interested contributing directly to the code base, please see CONTRIBUTING.md.

Acknowledgements

InterpretML was originally created by (equal contributions): Samuel Jenkins & Harsha Nori & Paul Koch & Rich Caruana

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

Citations

Paper link

Contact us

There are multiple ways to get in touch:

If a tree fell in your random forest, would anyone notice?

comments powered by Disqus