NEWS.md

In seagull: Lasso, Group Lasso, and Sparse-Group Lasso for Mixed Models

seagull 1.1.0

• Added new algorithm: fitted group lasso.

• Added new algorithm: fitted sparse-group lasso.

• Added new parameter: l2_fitted_values to enable the new algorithms.

• Added new parameter: step_size - mandatory for the fitted sparse-group lasso.

• Added new parameter: delta - mandatory for both new algorithms.

• Added new parameter: standardize to enable automated standardization of input data.

• Imported the R package "matrixStats" since it is used for the standardization.

• Updated the help pages and vignette to provide information about the new algorithms and parameters.

seagull 1.0.6

• Deleted "trace_progress = T" in all examples, since it caused --run-donttest to fail.

• Added a reference in the package documentation.

seagull 1.0.5

• Wrapped the entire example section in seagull.R into a \donttest environment to circumvent troubles with the Solaris OS.

seagull 1.0.4

• Replaced the pre-calculation of the matrix X^T X, by matrix-vector multiplications for the gradients in src\seagull_lasso.cpp, src\seagull_group_lasso.cpp, and src\seagull_sparse_group_lasso.cpp. This circumvents potential memory issues for the allocation of X^T X.

• Fixed an issue related to the variable TEMP2 in src\seagull_lasso.cpp, src\seagull_group_lasso.cpp, and src\seagull_sparse_group_lasso.cpp

seagull 1.0.3

• Replaced \dontrun by \donttest in R\seagull.R.

• Shortened title in DESCRIPTION.

seagull 1.0.2

• Fixed a limitation for the design matrix X. The matrix may now have more columns than rows. But each default algorithm to calculate max_lambda will fail, because the inverse of X^T X is explicitly needed. However, if a value for max_lambda is provided upon calling the function seagull, a solution will be calculated.

seagull 1.0.1

• Added parameter trace_progress. Default is FALSE.

• Added general vignette.

seagull 1.0.0

• Exchanged wrappers seagull_lasso, seagull_group_lasso, and seagull_sparse_group_lasso by seagull. The different penalties shall now be called by specifying the mixing parameter alpha. This parameter was initially only necessary for the sparse-group lasso. But the lasso and the group lasso are limiting cases, where alpha = 1 and alpha = 0, respectively. So, now both regularizations may be initialized by calling the function seagull with alpha = 1 or alpha = 0.

seagull 0.1.2

• Added documentation.

seagull 0.1.1

• Added different exemplary data set (seagull_data).

seagull 0.1.0

• Added wrapper seagull_lasso for the seagull_lasso_Rcpp.cpp

• Added wrapper seagull_group_lasso for the seagull_group_lasso_Rcpp.cpp

• Added wrapper seagull_sparse_group_lasso for the seagull_sparse_group_lasso_Rcpp.cpp

seagull documentation built on April 20, 2021, 5:06 p.m.