2018-11-10 Yang Feng <yang.feng@columbia.edu> Version 2.1.4
*improved the efficiency for computing Neyman-Pearson classifiers.
2018-02-13 Yang Feng <yang.feng@columbia.edu> Version 2.1.1
*added the slda classifier and the nonparametric naive bayes classifier
*added the adaptive split ratio option
2017-09-23 Yang Feng <yang.feng@columbia.edu> Version 2.1
*change the option of shiny for split numbers
2017-09-16 Yang Feng <yang.feng@columbia.edu> Version 2.0.9
*speed up the calculation for order statistics
*minor changes on plot legends and line width
2017-08-17 Yang Feng <yang.feng@columbia.edu> Version 2.0.8
*remove the custom option
*remove the nproc line, only keeps the band version
2017-02-13 Yang Feng <yang.feng@columbia.edu> Version 2.0.6
*add implementation for roc with cross-validation
2017-01-18 Yang Feng <yang.feng@columbia.edu> Version 2.0.5
*pass additional arguments to different classifiers via …
2017-01-11 Yang Feng <yang.feng@columbia.edu> Version 2.0.4
*improve the memory use for the outer function in find.order function
2016-09-18 Yang Feng <yang.feng@columbia.edu> Version 2.0.1
*using only the cutoffs from class 0 to create the NP-ROC band to match the theory.
2016-09-14 Yang Feng <yang.feng@columbia.edu> Version 2.0
*change the implementation of the interpolation for ROC confidence band by using a piecewise constant function to approximate the upper and lower bounds.
*using all the possible cutoff from class 1 and class 0 to create the NP-ROC band.
2016-09-08 Yang Feng <yang.feng@columbia.edu> Version 1.9
*change the implementation of the npc. In particular, originally, we are looking for the order statistic that satisfies the type I error upper bound with high-probability. Now, we investigate all possible cutoffs and evaluate its corresponding upper and lower bounds of type I and type II errors. This way, we do not need to impose the minimal sample size requirement.
*change the implementation of the ROC confidence band, here, the lower bound represents the upper bounds of type I error and 1-upper bounds of type II error. The upper bound represents the lower bounds of type I error and 1-lower bounds of type II error. Upon this revision, the comparison between two methods is more reasonable and has theoretical support.
2016-08-22 Yang Feng <yang.feng@columbia.edu> Version 1.4
*add the comparison between two NP-ROC curves
2016-08-15 Yang Feng <yang.feng@columbia.edu> Version 1.3
*fix the errors when the input x is a vector
2016-08-08 Yang Feng <yang.feng@columbia.edu> Version 1.2
*change the package title
*add the reference to the nproc paper in arxiv
2016-05-22 Yang Feng <yang.feng@columbia.edu> Version 1.0
*format the codes
*implement the adaptive split scheme, by dividing the alphalist into three regions
*change the custom method, into using all class 0 samples for determine the cutoff
*add options for the split proportion, whether to split adaptively
*change plot function to use dashed line if adaptive is TRUE for the small sample
size region.
2016-05-18 Yang Feng <yang.feng@columbia.edu> Version 0.8
*try out the implementation with adaptive choice of sample size for the order statistics
*change the order statistics selection using the exact formula
2016-05-14 Yang Feng <yang.feng@columbia.edu> Version 0.7
*change the parallel computing schedule to multiple splits
2016-05-13 Yang Feng <yang.feng@columbia.edu> Version 0.6
*change the bootstrap calculation to an explicit formula. The algorithm is more efficient now.
*change the starting point of the ROC curve to the alpha value for which it is possible to control the type I error with probability 1-delta
2016-05-08 Yang Feng <yang.feng@columbia.edu> Version 0.5
*change the mc.cores=1 in the Vignettes to be compatible with windows.
2016-04-18 Yang Feng <yang.feng@columbia.edu> Version 0.4
*change the default alphalist range.
2016-03-06 Yang Feng <yang.feng@columbia.edu> Version 0.3
*add ensemble implementation for npc, nproc.
2016-02-21 Yang Feng <yang.feng@columbia.edu> Version 0.2
* set random seed to 0 (customizable) for reproducibility in all functions
* add implementation of np-roc with pre-specified confidence level with conf parameter
* change SVM implementation to use raw scores instead of probability
* write a core function for nproc to speed up the process
* add functionality of running several classifiers at the same time for nproc
* add a plot function for nproc class, which can compare several classifiers
* added a vignette for a detailed demo of the package