II.
LibraryProcess overview
Reference · livelib-process:data-science-ml--model-interpretability
model-interpretability overview
Model Interpretability and Explainability Analysis - Comprehensive model interpretation pipeline with global and local explanations, feature importance analysis, SHAP/LIME integration, decision path analysis, counterfactual explanations, and bias detection with quality convergence gates.
Attributes
displayName
model-interpretability
description
Model Interpretability and Explainability Analysis - Comprehensive model interpretation pipeline with
global and local explanations, feature importance analysis, SHAP/LIME integration, decision path analysis,
counterfactual explanations, and bias detection with quality convergence gates.
libraryPath
library/specializations/data-science-ml/model-interpretability.js
specialization
data-science-ml
references
- - SHAP (SHapley Additive exPlanations): https://shap.readthedocs.io/ - LIME (Local Interpretable Model-agnostic Explanations): https://github.com/marcotcr/lime - InterpretML: https://interpret.ml/ - Alibi Explain: https://docs.seldon.io/projects/alibi/ - DALEX: https://dalex.drwhy.ai/ - What-If Tool: https://pair-code.github.io/what-if-tool/ - Captum (PyTorch): https://captum.ai/ - TensorFlow Model Analysis: https://www.tensorflow.org/tfx/guide/tfma
example
const result = await orchestrate('specializations/data-science-ml/model-interpretability', {
modelPath: 'models/churn-prediction-model.pkl',
modelType: 'classification',
dataPath: 'data/test-samples.csv',
targetColumn: 'churn',
explainabilityMethods: ['shap', 'lime', 'partial-dependence', 'counterfactual'],
targetQuality: 85,
maxIterations: 3
});
usesAgents
- general-purpose
Outgoing edges
lib_applies_to_domain1
- domain:data-science·DomainData Science
lib_belongs_to_specialization1
- specialization:data-science-ml·Specialization
lib_implements_workflow1
- workflow:data-pipeline-deployment·WorkflowData Pipeline Deployment
lib_involves_role1
- role:data-scientist·RoleData Scientist
Incoming edges
None.