CHANGES in `interval' VERSION 1.1-0.1
o fix vignette, put in proper directory
o fix getsurv to work in when times before first failure
o fix error in Aintmap for extreme case when intmap=(0,Inf) and k=1.
o added a modified bootstrap confidence interval for the NPMLE to the icfit function.
o change default for initfit in icfit, ictest, and wlr_trafo to NULL (avoids a rare problem with the initcomputeMLE function which could cause R to crash). Change help to explain this.
CHANGES in `interval' VERSION 1.1-0.0
o fix error in getsurv when the NPMLE is indeterminate in the interval from 0 to x. In that case getsurv could not calculate the interpolation.
o add conf.int option for icfit. This uses a modified boostrap confidence interval and can be very time consuming.
o change options for plot.icfit so that you can give par options (e.g., lwd, lty, col) for estimates (estpar) and confidence intervals (cipar).
o change '].icfit' to allow for confidence interval.
o update help files, add new references.
CHANGES in `interval' VERSION 1.0-1.2
o fix error in icfit when stratified data with at least one degenerate strata (i.e., NPMLE result for the degenerate strata is only one interval with pf=1 in that interval)
o fix error in icfit when data are degenerate such that all left=0 and right=Inf, now gives correct NPMLE
o remove non-ASCII characters (the i in naive) from Rnw file so that it passes checks.
CHANGES in `interval' VERSION 1.0-1.1
o change default for initfit in icfit, ictest, and wlr_trafo to NULL (avoids a rare problem with the initcomputeMLE function which could cause R to crash). Change help to explain this.
CHANGES in `interval' VERSION 1.0-1.0
o fix bug that gave error when analyzing right censored or exactly observed data. Needed to create Lin and Rin from within icfit before calling initcomputeMLE.
CHANGES in `interval' VERSION 1.0-0.0
o set tol=1e-09 in initcomputeMLE to avoid a very rare problem that occurred when calling computeMLE (version 0.1-3 of MLEcens).
Thanks to M. Maathuis for the suggestion.
o fix getsurv so that it works when strata is missing from the icfit object.
o fix [.icfit so that the attribute to the intmap is picked out when used
o fix ictest so that it does not use alternative="two.sidedAbs" internally when calling perm package, so now it does not give warnings
when calling perm package. Make the package require use of perm (>=1.0) for the same reason.
o make various changes to intervalPaper to meet JSS style.
o in ictest: change alternative="two.sided" so that it will not be confused with the method in coin. Now instead of having a separate option
for alternative called "two.sidedAbs", we drop that option and include another input, tsmethod, standing for two-sided method.
So these are the same:
Vers 0.9-9.6: alternative="two.sided"
Vers 1.0-0.0: alternative="two.sided" and tsmethod="central"
and these are the same:
Vers 0.9-9.6: alternative="two.sidedAbs"
Vers 1.0-0.0: alternative="two.sided" and tsmethod="abs"
The default is tsmethod="central". The structure here matches the exact2x2 and exactci packages.
CHANGES in `interval' VERSION 0.9-9.6
o fix problem with right censored data in icfit
- create Lin and Rin in SurvLR function
- in icfit.formula now use Lin and Rin in call to icfit.default
CHANGES in `interval' VERSION 0.9-9.5
o overall changes
- add more comments to code
- edit main vignette according to comments from reviewers
- make changes to help according to the changes outlined below, and suggestions from reviewers
o changes to icfit
- allow initfit to be the name of a function used to calculate the initfit
- make default initfit="initcomputeMLE", using the 'computeMLE' function from the MLEcens package.
This greatly increases the speed of the algorithm, but the original algorithm is kept.
Thus, we continue to do the checks on the Kuhn-Tucker conditions on the final estimate
and the initial fit need not be close to the final NPMLE.
- speed up each iteration of the EM algorithm by using matrix multiplication instead of
the apply function.
- output 'anypzero' as part of icfit objects created in icfit. The value 'anypzero' is a logical
to denote if any probability masses where set to zero. In other words, when TRUE
there are parameters on the boundary of the parameter space and the ad hoc adjustment to the score test
method is used (as had been used in the previous versions and detailed in the help) when method="scoretest".
- program restructured. Aintmap function now outside of icfit function
- if A matrix has 1 column, gives correct values for pf: pf==c(1)
o changes to wlr_trafo
- add method for data.frame so that it may be used easily with the coin package
- if length of pf=1, gives correct values for scores: all scores=0
o changes to ictest
- add scores="normal" and "general" option, these work only with permutation methods
- if length of pf=1 from overall fit, gives correct p-value: p.value=1
o changed getsurv to properly handle icfit objects that have more than one strata.
CHANGES in `interval' VERSION 0.7-5.5
o comment out extra code in icScoreTest that caused problems in Rcmd check
o change first parameter in wlr_trafo to x to remove Rcmd check problems