Skip to main content
TabPFN models are interpretable by design - their predictions are stable and reproducible thanks to in-context meta-learning. The Interpretability Extension adds SHAP (SHapley Additive exPlanations) support to quantify the contribution of each input feature to an individual prediction. Why it matters:
  • See which features drive model predictions.
  • Compare feature importance across samples.
  • Debug unexpected model behavior.
  • Communicate insights to stakeholders clearly and visually.

​
Shapley Values

SHAP values explain a single prediction by attributing the prediction’s deviation from the baseline (mean prediction) to individual features. They provide a consistent, game-theoretic measure of feature influence. Mathematically, each SHAP value represents the marginal contribution of a feature across all possible feature combinations.

​
Getting Started

Install the interpretability extension: 
pip install "tabpfn-extensions[interpretability]"
Then, use SHAP with any trained TabPFN model: 
from tabpfn import TabPFNClassifier
from tabpfn_extensions import interpretability
from sklearn.datasets import load_breast_cancer
from sklearn.model_selection import train_test_split

# Load dataset
data = load_breast_cancer()
X, y = data.data, data.target
feature_names = data.feature_names

# Train/test split
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.5, random_state=42)

# Train TabPFN model
clf = TabPFNClassifier()
clf.fit(X_train, y_train)

# Compute SHAP values for the first 50 samples
shap_values = interpretability.shap.get_shap_values(
    estimator=clf,
    test_x=X_test[:50],
    attribute_names=feature_names,
    algorithm="permutation",
)

# Visualize feature contributions
fig = interpretability.shap.plot_shap(shap_values)
SHAP values can be calculated for both classification and regression tasks. The extension supports CPU and GPU execution transparently.

​
Core Functions

FunctionDescription
get_shap_valuesCalculates SHAP values for the provided model and data subset.
plot_shapGenerates an interactive visualization showing feature contributions for each prediction.

​
Advanced Use Cases

  • Per-sample explainability: inspect individual predictions to see why the model produced a specific output.
  • Global interpretability: aggregate SHAP values across samples to rank features by importance.
  • Feature interaction analysis: identify features that amplify or offset contributions.
  • Model comparison: use SHAP to compare explanations between TabPFN variants.
⌘I