1.4
- added a formula interface through 'iCoxBoost'
- added generic function 'coef' for extracting estimated coefficients
- added a plot routine that provides coefficient paths
- added support for package 'parallel' (removing support for 'multicore' and older R versions)
- convergence problems for unpenalized covariates now are catched
1.3
- added option 'criterion' to allow for selection according to unpenalized scores
- added 'criterion="hpscore"' and 'criterion="hscore"' for heuristic evaluation
of only a subset of covariates in each boosting step
- Fixed a bug where results from "predict" without "newdata" and "linear.predictor"
in CoxBoost objects would have the wrong order (introduced in 1.2-1)
- added missing value check for covariate matrix
- implemented observation weights
1.2-2
- fixed a bug in the predict function ocurred when all coefficients
were equal to zero
- fixed bug where 'estimPVal' would with only one boosting step
- 'estimPVal' now also works for zero boosting steps
1.2-1
- improved speed of the core selection routine
- added faster code for the special case of binary covariate data
- added an option for not returning the matrix with the score statistics
for saving memory in applications with a huge number of covariates
- optimized memory usage for a large number of covariates
- covariates with standard deviation equal to zero now only are centered
- a matrix of the employed penalties know is only stored if the penalties,
changed. Otherwise the 'element' penalty is just a vector
- added support for 'multicore' package for cross-validation and p-value
estimation
- added an option for fitting on subsets of observations
- The coefficient matrix is now stored as a sparse matrix, employing
package 'Matrix'
- fixed the implementation of the p-value estimation
1.2
- added function 'estimPVal' for permutation-based p-value estimation
- improved the speed of the penalty updating code in PathBoost
1.1-1
- fixed bug in print method (introduced in 1.0-1) where the number of
non-zero coefficients would be taken from a wrong boosting step
1.1
- implemented penalty modification factors and penalty change distribution
via a connection matrix
- implemented estimation of models for competing risks
1.0-1
- implemented data adaptive rule for default penalty value
- fixed bug where output of the selected covariate would print the
wrong name in presence of unpenalized covariates
- Boosting now starts a step 0, i.e., also the model before updating
any of the coefficients of the penalized covariates is considered.
However, the unpenalized covariates will already have non-zero
values in boosting step 0.
This change breaks code that relies on the size of elements
"coefficients", "linear.predictors", or "Lambda" of CoxBoost objects
- implements parallel evaluation of cross-validation folds,
via package 'snowfall'
- speed improvements by replacing 'apply' and 'rbind' , most noticeably
for a large number of observations with a small number of covariates
1.0
* initial public release