Flexible Co-Data Learning for High-Dimensional Prediction
Obtain coefficients from 'ecpc' object
Create a list of constraints for co-data weight estimation
Create a group set (groups) of variables
Create a generalised penalty matrix
Create a co-data matrix Z for a group set
Create a co-data matrix Z of splines
Cross-validation for 'ecpc'
tools:::Rd_package_title("ecpc")
Fit adaptive multi-group ridge GLM with hypershrinkage
Fit hierarchical lasso using LOG penalty
Obtain hierarchy
Plot an 'ecpc' object
Perform posterior selection
Predict for new samples for `ecpc' object
Print summary of 'ecpc' object
Produce folds
Simulate data
Discretise continuous data in multiple granularities
Visualise a group set
Visualise estimated group set weights
Visualise estimated group weights
Fit linear, logistic and Cox survival regression models penalised with adaptive multi-group ridge penalties. The multi-group penalties correspond to groups of covariates defined by (multiple) co-data sources. Group hyperparameters are estimated with an empirical Bayes method of moments, penalised with an extra level of hyper shrinkage. Various types of hyper shrinkage may be used for various co-data. Co-data may be continuous or categorical. The method accommodates inclusion of unpenalised covariates, posterior selection of covariates and multiple data types. The model fit is used to predict for new samples. The name 'ecpc' stands for Empirical Bayes, Co-data learnt, Prediction and Covariate selection. See Van Nee et al. (2020) <arXiv:2005.04010>.