Compute SHAP values to attribute predictions to features across model types.
npx skills add https://github.com/xiechy/climate-ai --skill shap
Please help me install this Agent Skill. Skill: shap Source: https://github.com/xiechy/climate-ai/tree/main/scientific-packages/shap Command: npx skills add https://github.com/xiechy/climate-ai --skill shap
Machine learning models are often "black boxes," making it difficult to understand why they make certain predictions. This skill provides a unified, theoretically sound approach to explain model outputs, enabling users to interpret feature importance, debug model behavior, and ensure fairness.
To explain an XGBoost model, first train your model, then:
import shap
explainer = shap.TreeExplainer(model)
shap_values = explainer(X_test)
shap.plots.beeswarm(shap_values)
High-intent search queries and answers about installing and using this skill.
SHAP values quantify each feature's contribution to a prediction by computing game-theoretic values. Train your model, initialize a SHAP explainer (TreeExplainer for tree-based models, DeepExplainer for neural networks, KernelExplainer for black-box models), call it on your test data, then visualize results using waterfall, beeswarm, or force plots to see which features pushed predictions in which direction.
Yes. SHAP supports tree-based ensembles (XGBoost, LightGBM, CatBoost, Random Forest), deep learning frameworks (TensorFlow, PyTorch, Keras), linear models, and black-box models through specialized explainers that compute SHAP values efficiently for each architecture.
Waterfall plots show how features push individual predictions away from the base value; beeswarm plots reveal global feature importance across all samples; bar charts rank features by average impact; scatter plots explore feature-value relationships; force plots display prediction breakdown compactly; heatmaps show interactions. Choose based on whether you're explaining one prediction or analyzing patterns across many.
Yes. SHAP values expose which features disproportionately influence predictions, helping identify bias in lending, hiring, or healthcare models. Generate SHAP values across demographic groups and compare feature contributions to spot fairness violations and ensure equitable decision-making.
You need a trained model and test or evaluation data in the same format the model expects. For TreeExplainer and some explainers, provide background data (a sample of training data) to establish baseline expectations; for model-agnostic explainers like KernelExplainer, background data helps approximate model behavior efficiently.
Compute SHAP values for competing models on the same test set, then compare feature importance distributions, interaction patterns, and individual prediction breakdowns. Models with clearer, more consistent feature contributions and fewer spurious interactions typically generalize better and are easier to trust in production.
gracefullightExplain machine learning model predictions using SHAP values and plots.
xiechyCompute SHAP values to attribute predictions to features across model types.
DTMC-marketplaceValidate AI models and datasets for bias and fairness using Python scripts.
tondevrelAnalyze scikit-learn model feature importance with SHAP, LIME, and partial dependence plots.
