This is a simple adaptation of the getOptcv.glmnet function from glmnet.
The code takes a grid of lambda penalties along with a vector of associated mean squared errors (mse), standard errors (se), and the mean squared error from each cross-validation run (fullMSE).
From this three potential options for the cross-validated penalty parameter are computed. 1) The lambda that has the minimum average mean squared error across all the cross-validation runs (lambda$lambda.min),
2) The lambda the largest lambda that is associated with an average cross-validated mean squared error within one standard error of the minimum average cross-validated mean squared error (lambda$lambda.1se), and
3) the lamda that is the median of the set of lambdas (lambda$lambda.median).
getOptcv.scul(lambdapath, mse, se, fullMSE)| lambdapath | A grid of lambdas that is used in each cross-validation run as potential options for the optimal penalty parameter. |
|---|---|
| mse | A vector of the average mean squared error (average across cross-validation runs) for each given |
| se | A vector of the standard error associated with each average mean squared error (average across cross-validation runs) for each given |
| fullMSE | A matrix of the mean squared error for each cross-validation run and each given |