LIME Interpretability

docs/api/interpret.rst

@@ -50,6 +50,16 @@ New to interpretability in PyHealth? Check out these complete examples:
   - Compare different baseline strategies for background sample generation
   - Decode attributions to human-readable medical codes and lab measurements
 
+**LIME Example:**
+
+- ``examples/lime_stagenet_mimic4.py`` - Demonstrates LIME (Local Interpretable Model-agnostic Explanations) for StageNet mortality prediction. Shows how to:
+
+  - Compute local linear approximations to explain model predictions
+  - Generate perturbations around input samples to train interpretable models
+  - Compare different regularization methods (Lasso vs Ridge) for feature selection
+  - Test various distance kernels (cosine vs euclidean) and sample sizes
+  - Decode attributions to human-readable medical codes and lab measurements
+
 These examples provide end-to-end workflows from loading data to interpreting and evaluating attributions.
 
 Available Methods
@@ -64,4 +74,5 @@ Available Methods
     interpret/pyhealth.interpret.methods.deeplift
     interpret/pyhealth.interpret.methods.integrated_gradients
     interpret/pyhealth.interpret.methods.shap
+    interpret/pyhealth.interpret.methods.lime
 

docs/api/interpret/pyhealth.interpret.methods.lime.rst

@@ -0,0 +1,19 @@
+pyhealth.interpret.methods.lime
+===============================
+
+Overview
+--------
+
+:class:`pyhealth.interpret.methods.lime.LimeExplainer` provides LIME (Local Interpretable Model-agnostic Explanations)
+attributions for PyHealth models by approximating the model locally with an interpretable linear model. Consult the class docstring for 
+detailed guidance, usage notes, and examples. A full workflow is demonstrated in
+``examples/lime_stagenet_mimic4.py``.
+
+API Reference
+-------------
+
+.. autoclass:: pyhealth.interpret.methods.lime.LimeExplainer
+    :members:
+    :undoc-members:
+    :show-inheritance:
+    :member-order: bysource