Interpretable machine learning models (imodels) 🔍
Python package for concise, transparent, and accurate predictive modeling. All sklearn-compatible and easily understandable. Pull requests very welcome!
Docs • Popular imodels • Custom imodels • Demo notebooks
Popular interpretable models
Implementations of different interpretable models, all compatible with scikit-learn. The interpretable models can be easily used and installed:
from imodels import BayesianRuleListClassifier, GreedyRuleListClassifier, SkopeRulesClassifier, IRFClassifier from imodels import SLIMRegressor, RuleFitRegressor model = BayesianRuleListClassifier() # initialize a model model.fit(X_train, y_train) # fit model preds = model.predict(X_test) # discrete predictions: shape is (n_test, 1) preds_proba = model.predict_proba(X_test) # predicted probabilities: shape is (n_test, n_classes)
pip install git+https://github.com/csinva/imodels (see here for help). Contains the following models:
|Bayesian rule list||🗂️, 🔗, 📄||Learns a compact rule list by sampling rule lists (rather than using a greedy heuristic)|
|Rulefit||🗂️, 🔗, 📄||Extracts rules from a decision tree then builds a sparse linear model with them|
|Skope rules||🗂️, 🔗||Extracts rules from gradient-boosted trees, deduplicates them, then forms a linear combination of them based on their OOB precision|
|Sparse integer linear model||🗂️, 📄||Forces coefficients to be integers|
|Greedy rule list||🗂️, 🔗||Uses CART to learn a list (only a single path), rather than a decision tree|
|Iterative random forest||🗂️, 🔗, 📄||(In progress) Repeatedly fit random forest, giving features with high importance a higher chance of being selected.|
|Optimal classification tree||🗂️, 🔗, 📄||(In progress) Learns succinct trees using global optimization rather than greedy heuristics|
|Rule sets||(Coming soon) Many popular rule sets including SLIPPER, Lightweight Rule Induction, MLRules|
Reference includes docs 🗂️, reference code implementation 🔗, and research paper 📄
Custom interpretable models
The code here contains many useful and readable functions for a variety of rule-based models, contained in the util folder. This includes functions and simple classes for rule deduplication, rule screening, converting between trees, rulesets, and pytorch neural nets. The final derived rules easily allows for extending any of the following general classes of models:
|Rule set||Rule list||(Decision) Rule tree||Algebraic models|
Demos are contained in the notebooks folder.
- model_based.ipynb, demos the imodels package. It shows how to fit, predict, and visualize with different interpretable models
- this notebook shows an example of using
imodelsfor deriving a clinical decision rule
- After fitting models, we can also do posthoc analysis, following the cheat-sheet below
- posthoc.ipynb - shows different simple analyses to interpret a trained model
- uncertainty.ipynb - basic code to get uncertainty estimates for a model
- Interpretable ML good quick overview: (murdoch et al. 2019, pdf)
- Interpretable ML book: (molnar 2019, pdf)
- Case for interpretable models rather than post-hoc explanation: (rudin 2019, pdf) - good explanation of why one should use interpretable models
- Review on evaluating interpretability (doshi-velez & kim 2017, pdf)
- Reference implementations (also linked above): the code here heavily derives from (and in some case is just a wrapper for) the wonderful work of previous projects. We seek to to extract out, combine, and maintain select relevant parts of these projects.
- Related packages
For updates, star the repo, see this related repo, or follow @csinva_. Please make sure to give authors of original methods / base implementations appropriate credit!
Expand source code
""" .. include:: ../readme.md """ # Python `imodels` package for interpretable models compatible with scikit-learn. # Github repo available [here](https://github.com/csinva/interpretability-implementations-demos). from .bayesian_rule_list.bayesian_rule_list import BayesianRuleListClassifier from .greedy_rule_list import GreedyRuleListClassifier from .iterative_random_forest.iterative_random_forest import IRFClassifier from .rule_fit import RuleFitRegressor from .skope_rules import SkopeRulesClassifier from .slim import SLIMRegressor CLASSIFIERS = BayesianRuleListClassifier, GreedyRuleListClassifier, IRFClassifier REGRESSORS = RuleFitRegressor, SkopeRulesClassifier, SLIMRegressor # from .optimal_classification_tree import OptimalTreeModel
bayesian rule list (based on this implementation) - learn a compact …
greedy rule list (based on this implementation). …
Repeatedly fit random forest, giving features with high importance a higher chance of being selected.
Still in progress - should optimize globally over all possible splits rather than use greedy 1-by-1 splits
Linear model of tree-based decision rules …
Code adapted with only minor changes from here. Full credit to the authors …
wrapper for sparse, integer linear models …
Shared utilities for implementing different interpretable models.