SHAP-IQ: SHAP Interaction Quantification

An interaction may speak more than a thousand main effects.

❗ Note: This application is still in alpha and still under active development. ❗

SHAP Interaction Quantification (short SHAP-IQ) is an XAI framework extending on the well-known shap explanations by introducing interactions to the equation. Shapley interactions extend on indivdual Shapley values by quantifying the synergy effect between machine learning entities such as features, data points, or weak learners in ensemble models. Synergies between these entities (also called players in game theory jargon) allows for a more intricate evaluation of your black-box models!

🛠️ Install

shapiq is intended to work with Python 3.9 and above. Installation can be done via pip:

pip install shapiq

⭐ Quickstart

You can use shapiq in different ways. If you have a trained model you can rely on the shapiq.explainer classes. If you are interested in the underlying game theoretic algorithms, then check out the shapiq.approximator modules. You can also plot and visualize your interaction scores with shapiq.plot.

📈 Compute k-SII values

Explain your models with Shapley interaction values like the k-SII values:

# train a modelfromsklearn.ensembleimportRandomForestRegressormodel=RandomForestRegressor(n_estimators=50, random_state=42)
model.fit(x_train, y_train)
# explain with k-SII interaction scoresfromshapiqimportInteractionExplainerexplainer=InteractionExplainer(
model=model.predict,
background_data=x_train,
index="k-SII",
max_order=2
)
interaction_values=explainer.explain(x_explain, budget=2000)
print(interaction_values)
>>>InteractionValues(
>>>index=k-SII, max_order=2, min_order=1, estimated=True, estimation_budget=2000,
>>>values={
>>> (0,): -91.0403, # main effect for feature 0>>> (1,): 4.1264, # main effect for feature 1>>> (2,): -0.4724, # main effect for feature 2>>> ...
>>> (0, 1): -0.8073, # 2-way interaction for feature 0 and 1>>> (0, 2): 2.469, # 2-way interaction for feature 0 and 2>>> ...
>>> (10, 11): 0.4057# 2-way interaction for feature 10 and 11>>> }
>>> )

📊 Visualize your Interactions

One handy way of visualizing interaction scores (up to order 2) are network plots. You can see an example of such a plot below. The nodes represent attribution scores and the edges represent the interactions. The strength and size of the nodes and edges are proportional to the absolute value of the attribution scores and interaction scores, respectively.

fromshapiq.plotimportnetwork_plotnetwork_plot(
first_order_values=k_sii_first_order, # first order k-SII valuessecond_order_values=k_sii_second_order# second order k-SII values
)

The pseudo-code above can produce the following plot (here also an image is added):

📖 Documentation

The documentation for shapiq can be found here.

💬 Citation

If you ejnoyshapiq consider starring ⭐ the repository. If you really enjoy the package or it has been useful to you, and you would like to cite it in a scientific publication, please refer to the paper accepted at NeurIPS'23:

@article{shapiq,
author = {Fabian Fumagalli and Maximilian Muschalik and Patrick Kolpaczki and Eyke H{\"{u}}llermeier and Barbara Hammer},
title = {{SHAP-IQ:} Unified Approximation of any-order Shapley Interactions},
journal = {CoRR},
volume = {abs/2303.01179},
year = {2023},
doi = {10.48550/ARXIV.2303.01179},
eprinttype = {arXiv}
}