NEWS

==========================================
modEvA v2.0 (committed 2020-01-21)
==========================================

## Modified functions:

* modEvA-internal
- removed .Random.seed

* multModEv, modEvAmethods, modEvA-internal
- added "MeanPrecision", "AUCPR" and "F1score"


## Other modified files:

* multModEv.Rd
- added "AUCPR" to one of the examples


==========================================
modEvA v2.0 (committed 2020-01-20)
==========================================

## Modified functions:

* AUC
- added 'meanPrecision' to returned list (if simplif = FALSE)
- "AUC[PR]" mentioned on plot when curve = "PR"

* optiThresh
- added "F1score" to goodness.measures in optimals.each
- added Liu et al. 2005 reference to threshold criteria

* threshMeasures, modEvAmethods
- added "Precision" and "Recall" (even though synonyms with "PPP" and "Sensitivity")

* optiPair
- changed pch of measure 1 to 19 (slightly larger than pch 20) so it's still visible when overlapped by measure 2

* multModEv
- added 'plot = FALSE' to internal AUC call


## Other modified files:

* threshMeasures.Rd, optiThresh.Rd, optiPair.Rd
- added "Note" saying sens=recall and PPP=precision, and defining F1score 


==========================================
modEvA v1.8 (committed 2020-01-17)
==========================================

## Modified functions:

* AUC
- removed artificial zeros (instead of NaN) when curve = "PR"
- reversed the reference diagonal if curve = "PR"
- value now placed higher on the plot if curve = "PR" to avoid superposition


## Other changes:

- added 'inst' folder with article citation information


==========================================
modEvA v1.7 (committed 2020-01-16)
==========================================

## Modified functions:

* AUC
- implemented methods "trapezoid" and "integrate"
- added area calculation for "PR" curve


* threshMeasures, evaluate, modEvAmethods, (optiThresh, optiPair)
- added "F1score" to available 'measures'


==========================================
modEvA v1.5 (committed 2020-01-15)
==========================================

## New functions:

* predDensity


## Modified functions:

* AUC
- new 'method' argument, which in the future can be "rank", "trapezoid" or "integrate" (currently only "rank" is implemented)
- new 'curve' argument which can be "ROC" or "PR" (precision-recall)
- arguments 'roc.col', 'roc.lty' and 'roc.lwd' renamed to 'curve.col', 'curve.lty' and 'curve.lwd'
- 'xlab' and 'ylab' now default to "auto", to be generated according to 'curve'
- removed 'main' argument (can be provided via '...')
- 'plot.preds' can now be either a logical value as before, or a character "curve" and/or "bottom"
- 'plot.preds' circles now plotted in darkgrey
- title in .Rd file changed to "Area Under the Curve" (removed "ROC")


## Other changes:

- corrected some problems in .Rd files, to reduce LaTeX errors when creating PDF version of manual.


==========================================
modEvA v1.4.2 (committed 2020-01-03)
==========================================

## Modified functions:

* plotGLM
- corrected to adj=1 when plotting values on the right of the plot


==========================================
modEvA v1.4.1 (committed 2019-10-18)
==========================================

## Modified functions:

* RsqGLM
- added NA handling for 'cor'


==========================================
modEvA v1.4 (committed 2018-11-28)
==========================================

## Modified functions:

* AUC
- added arguments diag, diag.col, diag.lty, roc.col, roc.lty, roc.lwd
- 'simplif = TRUE' no longer overrides 'plot = TRUE'

* MillerCalib
- added arguments diag, line.col


## Other modified files:

- edited .Rd files (AUC, HLfit, evaluate, optiPair, threshMeasures, modEvA-package) to reflect difference between discrimination and classification
- updated index.php


==========================================
modEvA v1.3.3 (committed 2017-03-24)
==========================================

## Modified functions:

* varPart:
- bug fixed (with na.omit) when only two factors (A, B, AB)


==========================================
modEvA v1.3.2 (committed 2016-07-12)
==========================================

## Modified functions:

* getBins (affecting HLfit):
- bug fixed in "bin.method = size.bins", which was giving two different messages regarding ignored arguments

* HLfit:
- help file slightly clarified


==========================================
modEvA v1.3.1 (committed 2016-06-27)
==========================================

## Modified functions:

* getBins (affecting HLfit):
- bug fixed in "bin.method = quantiles", now providing even-sized bins (=SPSS)
- added argument 'quantile.type', to pass to 'quantile' function

* HLfit:
- added argument 'quantile.type', to pass to 'quantile' function
- added argument 'verbosity'

* multModEv:
- removed default 'thresh' (must be user-specified)
- replaced argument 'quiet' with 'verbosity' for coherence with other functions

* threshMeasures:
- removed default 'thresh' (must be user-specified)
- replaced argument 'messages' with 'verbosity' for coherence with other functions


==========================================
modEvA v1.3 (committed 2016-06-16)
==========================================

## Modified functions:

* getBins (affecting HLfit):
- bug fixed in "bin.method = quantiles", removing last additional bin of size 1
- help file improved to better explain what each bin.method does and what arguments it ignores
- added argument 'verbosity'

* HLfit:
- added examples to help file to illustrate differences in bin.methods


==========================================
modEvA v1.2.9 (committed 2016-06-06)
==========================================

## Modified functions:

* getBins (affecting HLfit):
- bug fixed in "bin.method = n.bins, fixed.bin.size = TRUE"
- help file updated to explain what each bin.method does
- removed default bin.method (must be specified by user)

* multModEv:
- included '...' for additional arguments to pass to HLfit
- removed default bin.method (must be specified if HL in measures)

* modEvAmethods:
- excluded "Miller.p" (previously eliminated from MillerCalib)


==========================================
modEvA v1.2.8 (committed 2016-04-18)
==========================================

## Modified functions:

* varPart:
- 'plot.unexpl' now also rounded to 'plot.digits'
- 'model.type' deprecated (message emitted)
- AB or ABC now also mandatory, and unexplained var 0, for GLMs
- help file updated

* plotGLM:
- warning on preds outside [0,1] now only emitted when appropriate

* AUC, getBins, HLfit, MillerCalib, multModEv, optiPair, optiThresh, plotGLM, threshMeasures:
- error when model(s) provided but not binomial logit glm


==========================================
modEvA v1.2.7 (committed 2016-04-15)
==========================================

## Modified functions:

* varPart:
- added argument 'plot.unexpl = TRUE'
- explicit error message for 3-factor LM with no ABC
- help file now with both LM and GLM examples
- help file now notes that wrong input = wrong result


==========================================
modEvA v1.2.6 (committed 2016-04-13)
==========================================

## Modified functions:

* AUC:
- error replaced with warning when pred values outside the [0,1] interval
- added argument plot.digits = 3 (for plot.values)

* varPart:
- added arguments 'main' and 'cex.main'
- slightly increased default cex values
- updated help file to make GLM use a bit clearer

* plotGLM:
- error replaced with warning when pred values outside the [0,1] interval
- added pseudo-R-squared measures to plot.values
- changed plot.values default to TRUE
- added argument plot.digits = 3 (for plot.values)


==========================================
modEvA v1.2.5 (committed 2016-03-30)
==========================================

## Modified functions:

* getBins:
- eliminated error when pred values outside the [0,1] interval

* HLfit:
- error replaced with warning on pred values outside the [0,1] interval
- plot now stretches if pred values exceed [0,1] interval

* MillerCalib:
- error replaced with warning on pred values outside the [0,1] interval
- default digits (for plot.values) lowered to 2
- slope p-value removed (values looked wrong)
- NaNs avoided by converting 0 and 1 in 'pred' to (1 -) 2e-16
- increased plot ymax by 0.2
- changed plot text location to bottom right


==========================================
modEvA v1.2.4 (committed 2016-03-22)
==========================================

## Modified functions:

* multModEv:
- warning now emitted, and help file updated, about calibration measures being valid only for probability (when input is obs.data + pred.data)


==========================================
modEvA v1.2.3 (committed 2015-12-10)
==========================================

## Modified functions:

* prevalence:
- 'na.rm = TRUE' added


==========================================
modEvA v1.2.2 (committed 2015-11-27)
==========================================

## Modified functions:

* Dsquared, RsqGLM, plotGLM:
- NaN-caused errors avoided by converting 0 and 1 in 'pred' to (1 -) 2e-16


==========================================
modEvA v1.2.1 (committed 2015-11-26)
==========================================

## Modified functions:

* Dsquared, RsqGLM:
- NaN-caused errors avoided by converting 0 in 'pred' to smallest computable positive number


==========================================
modEvA v1.2
==========================================

## Modified functions:

* AUC, Dsquared, RsqGLM, threshMeasures, optiPair, HLfit & MillerCalib now omit NAs

* Dsquared based on obs&pred now available also for Poisson GLMs