explainiverse 0.3.0__py3-none-any.whl → 0.5.0__py3-none-any.whl

explainiverse/__init__.py

@@ -33,7 +33,7 @@ from explainiverse.adapters.sklearn_adapter import SklearnAdapter
 from explainiverse.adapters import TORCH_AVAILABLE
 from explainiverse.engine.suite import ExplanationSuite
 
-__version__ = "0.2.5"
+__version__ = "0.5.0"
 
 __all__ = [
     # Core

explainiverse/core/registry.py

@@ -372,6 +372,8 @@ def _create_default_registry() -> ExplainerRegistry:
     from explainiverse.explainers.gradient.integrated_gradients import IntegratedGradientsExplainer
     from explainiverse.explainers.gradient.gradcam import GradCAMExplainer
     from explainiverse.explainers.gradient.deeplift import DeepLIFTExplainer, DeepLIFTShapExplainer
+    from explainiverse.explainers.gradient.smoothgrad import SmoothGradExplainer
+    from explainiverse.explainers.example_based.protodash import ProtoDashExplainer
     
     registry = ExplainerRegistry()
     
@@ -532,6 +534,23 @@ def _create_default_registry() -> ExplainerRegistry:
         )
     )
     
+    # Register SmoothGrad (for neural networks)
+    registry.register(
+        name="smoothgrad",
+        explainer_class=SmoothGradExplainer,
+        meta=ExplainerMeta(
+            scope="local",
+            model_types=["neural"],
+            data_types=["tabular", "image"],
+            task_types=["classification", "regression"],
+            description="SmoothGrad - noise-averaged gradients for smoother saliency maps (requires PyTorch)",
+            paper_reference="Smilkov et al., 2017 - 'SmoothGrad: removing noise by adding noise' (ICML Workshop)",
+            complexity="O(n_samples * forward_pass)",
+            requires_training_data=False,
+            supports_batching=True
+        )
+    )
+    
     # =========================================================================
     # Global Explainers (model-level)
     # =========================================================================
@@ -604,6 +623,27 @@ def _create_default_registry() -> ExplainerRegistry:
         )
     )
     
+    # =========================================================================
+    # Example-Based Explainers
+    # =========================================================================
+    
+    # Register ProtoDash
+    registry.register(
+        name="protodash",
+        explainer_class=ProtoDashExplainer,
+        meta=ExplainerMeta(
+            scope="local",
+            model_types=["any"],
+            data_types=["tabular"],
+            task_types=["classification", "regression"],
+            description="ProtoDash - prototype selection with importance weights for example-based explanations",
+            paper_reference="Gurumoorthy et al., 2019 - 'Efficient Data Representation by Selecting Prototypes' (ICDM)",
+            complexity="O(n_prototypes * n_samples^2)",
+            requires_training_data=True,
+            supports_batching=True
+        )
+    )
+    
     return registry
 
 

explainiverse/explainers/__init__.py

@@ -9,12 +9,17 @@ Local Explainers (instance-level):
 - Anchors: High-precision rule-based explanations
 - Counterfactual: Diverse counterfactual explanations
 - Integrated Gradients: Gradient-based attributions for neural networks
+- DeepLIFT: Reference-based attributions for neural networks
+- DeepSHAP: DeepLIFT combined with SHAP for neural networks
 
 Global Explainers (model-level):
 - Permutation Importance: Feature importance via permutation
 - Partial Dependence: Marginal feature effects (PDP)
 - ALE: Accumulated Local Effects (unbiased for correlated features)
 - SAGE: Shapley Additive Global importancE
+
+Example-Based Explainers:
+- ProtoDash: Prototype selection with importance weights
 """
 
 from explainiverse.explainers.attribution.lime_wrapper import LimeExplainer
@@ -29,6 +34,7 @@ from explainiverse.explainers.global_explainers.sage import SAGEExplainer
 from explainiverse.explainers.gradient.integrated_gradients import IntegratedGradientsExplainer
 from explainiverse.explainers.gradient.gradcam import GradCAMExplainer
 from explainiverse.explainers.gradient.deeplift import DeepLIFTExplainer, DeepLIFTShapExplainer
+from explainiverse.explainers.example_based.protodash import ProtoDashExplainer
 
 __all__ = [
     # Local explainers
@@ -46,4 +52,6 @@ __all__ = [
     "PartialDependenceExplainer",
     "ALEExplainer",
     "SAGEExplainer",
+    # Example-based explainers
+    "ProtoDashExplainer",
 ]

explainiverse/explainers/example_based/__init__.py

@@ -0,0 +1,18 @@
+# src/explainiverse/explainers/example_based/__init__.py
+"""
+Example-based explanation methods.
+
+These methods explain models by identifying representative examples
+from the training data, rather than computing feature attributions.
+
+Methods:
+- ProtoDash: Select prototypical examples with importance weights
+- (Future) Influence Functions: Identify training examples that most affect predictions
+- (Future) MMD-Critic: Find prototypes and criticisms
+"""
+
+from explainiverse.explainers.example_based.protodash import ProtoDashExplainer
+
+__all__ = [
+    "ProtoDashExplainer",
+]