sparsesurv

sparsesurv fits sparse survival models via knowledge distillation, providing an sklearn-like API to mitigate regularization hyperparameter sensitivity in high-dimensional survival analysis.

Key Features:

• Knowledge Distillation Framework: Transfers knowledge from complex teacher models to simpler student models to learn sparse survival representations.
• Teacher-Student Model Pairs: Includes the semi-parametric accelerated failure time model and the extended hazards model as teachers, which lack existing Python implementations.
• In-House Survival Function Estimators: Provides built-in survival function estimation without requiring external packages.
• Regularization Path Stability: Achieves competitive discriminative performance relative to glmnet while reducing sensitivity to regularization hyperparameter choice.
• sklearn-Compatible API: Offers an interface consistent with scikit-learn conventions for integration into Python machine learning workflows.

Scientific Applications:

• High-Dimensional Survival Analysis: Enables variable selection and survival modeling for datasets with many predictor variables relative to observations.
• Model Interpretability: Produces sparse models with limited non-zero coefficients to improve interpretability and transportability of survival predictions.
• Regularization Hyperparameter Optimization: Simplifies the selection of regularization parameters compared to traditional Lasso-regularized Cox proportional hazards models.
• Cross-Platform Validation: Benchmarked against R-based Elastic Net regularized linear Cox proportional hazards models from the glmnet package.

Methodology:

sparsesurv employs knowledge distillation where complex teacher models (semi-parametric accelerated failure time and extended hazards models) transfer knowledge to simpler student models, with in-house survival function estimators and validation against glmnet-implemented Elastic Net regularized linear Cox proportional hazards models.