============================================== modEvA v3.13.3 (committed 2024-01-19) -> CRAN ============================================== ## Modified functions: * threshMeasures - set plot=FALSE on confusionMatrix() call * confusionMatrix - set default 'plot' to FALSE (for back-compatibility) ============================================ modEvA v3.13.2 (committed 2024-01-18) ============================================ ## Modified functions: * confusionMatrix - add arguments plot=TRUE, classes=FALSE and ... ============================================ modEvA v3.13.1 (committed 2024-01-17) ============================================ ## Modified functions: * confusionLabel - assign consistent colours to output raster categories - add plot=TRUE and ... arguments ============================================ modEvA v3.13 (committed 2024-01-16) ============================================ ## Modified functions: * confusionLabel - output SpatRaster (if input 'pred' is SpatRaster) now properly categorical ## Other modified files: * AUC.Rd - unimplemented argument 'FPR.limits' now suggests using 'pROC::roc' ============================================ modEvA v3.12 (committed 2023-12-14) ============================================ ## Modified functions: * applyThreshold, inputMunch - added 'verbosity' argument to pass to ptsrast2obspred() * AUC - added 'grid.lty' argument (used e.g. in modevapp) - 'grid' now at 0.1 (not threshold) intervals * getThreshold - finished implementing 'maxF' - implemented maxJaccard, maxSorensen (= maxF) * optiThresh - replaced single with double ||, && - added modEvAmethods("similarity") to default 'measures' - added maxJaccard and maxSorensen criteria * modEvAmethods - added Jaccard, Sorensen and their maximizations * arrangePlots - return c(1, 1) if n.plots <= 1 ## Other modified files: * Boyce.Rd - 'Note' now mentions that Boyce is designed for suitability or favourability, not probability * similarity.Rd - 'Description' and 'Details' now mention equivalence to Li & Guo's F-measures ============================================ modEvA v3.11 (committed 2023-11-25) -> CRAN ============================================ ## Modified functions: * Dsquared - added dismo.version argument: if FALSE, uses the previous version of Dsquared, as the dismo version sometimes produces negative or NaN results ============================================ modEvA v3.10 (committed 2023-11-10) -> CRAN ============================================ ## New functions: * quantReclass * similarity ## Modified functions: * modEvAmethods - added methods for similarity() ============================================ modEvA v3.9.6 (committed 2023-10-04) ============================================ ## Modified functions: * Dsquared - warning emitted and Dsquared calculated correctly even when family binomial and response not binary (following bug report by Elic M. Weitzel) * varImp - metric reported in console result - absolute (rather than original) values now returned for z value (if model is glm) * AUC - added 'GiniCoefficient' to outputs when simplif=FALSE ============================================ modEvA v3.9.5 (committed 2023-01-02) ============================================ ## Modified functions: * varImp - barplot now also gets error bars replotted on top of points if plot.points=TRUE - implemented for models of class "lbart" and "pbart" of pkg 'BART' (though not ideal for group.cats = TRUE) - group.cats default now FALSE - started implementation for flexBART models (still imperfect because object has no var names; issue submitted at https://github.com/skdeshpande91/flexBART/issues/8) ============================================ modEvA v3.9.4 (committed 2023-04-26) ============================================ ## Modified functions: * varImp - new argument group.cats=TRUE, for BART models, to sum up the contributions of (one-hot encoded) categorical variable levels ============================================ modEvA v3.9.3 (committed 2023-04-14) -> CRAN ============================================ ## Modified functions: * AUC - 'method' default now NULL, to avoid unsolicited warning when curve="PR" because default method was "rank" ## Other modified files: * varImp.Rd - improved info on parameters for different plot types (e.g. cex.axis for lollipop, cex.names for barplot) ============================================ modEvA v3.9.2 (committed 2023-03-19) ============================================ ## Modified functions: * Boyce - explicit warning instead of obscure error if 'obs' has no presences - warning also when 'obs' has only presences (Boyce = NA) * optiThresh - explicit warning (and results NA) instead of obscure error if 'obs' has either no presences or no absences - empty plot instead of error for incalculable metrics * threshMeasures - explicit warning instead of obscure error if 'obs' has either no presences or no absences ============================================ modEvA v3.9.1 (committed 2023-02-26) ============================================ ## Modified functions: * varImp - error if family is not 'binomial', to avoid obscure error otherwise because no 'z-value' column in summary * confusionMatrix, confusionLabel - changed defaults 'interval = interval, quant = quant' to 'interval = 0.01, quant = 0' ## Other modified files: * CITATION - replaced old-style personList() with c(), and citEntry() with bibentry(), as per new CRAN requirements ============================================ modEvA v3.9 (committed 2023-01-12) -> CRAN ============================================ ## Modified functions: * varImp - corrected point locations (accounting for space between bars) for barplot - error.bars in lollipop chart re-plotted on top to improve visibility - help file cites Weissgerber et al. * predDensity - 'ci' now plotted as semi-transparent rectangle (rather than just vertical ablines) - added 'ci' example to help file ============================================ modEvA v3.8.9 (committed 2023-01-11) ============================================ ## Modified functions: * varImp - implemented plot.points argument - arguments 'ci' and 'ci.type' replaced by argument 'error.bars' - 'error.bars' can also be a numeric for a x% confidence interval - added 'grid' argument - added "imp.type" argument (placeholder, so far only option "each") - set 'border' to NA in legend and barplot - fixed legend colour bug - added check for loaded 'gbm' when needed * predDensity - added 'ci' argument - added '...' for arguments to pass to 'hist' ============================================ modEvA v3.8.8 (committed 2023-01-05) ============================================ ## New functions: * RMSE * lollipop ## Modified functions: * varImp - added "ylim" argument (default "auto"), and xpd=FALSE to 'barplot' call - changed default positive colour for better contrast with CI lines in barplot - new default plot.type = "lollipop" * threshMeasures - "OddsRatio" removed from default 'measures' - 'ylim' argument removed (default 0,1 not needed after OddsRation removal; 'ylim' can still be used under '...') - new argument 'plot.type' - new default plot.type = "lollipop" * RsqGLM - new argument 'plot.type' - new default plot.type = "lollipop" ============================================ modEvA v3.8.7 (committed 2022-12-22) ============================================ ## New functions: * varImp ## Modified functions: * mod2obspred - added supressMessages() when predicting with GBM ============================================ modEvA v3.8.6 (committed 2022-12-20) ============================================ ## Modified functions: * Dsquared - 'family' is now guessed from data (with message) if null - extended to models other than GLM ============================================ modEvA v3.8.5 (committed 2022-11-06) ============================================ ## Modified functions: * ptsrast2obspred - added (or 'rm.dup.points=TRUE' for 'Boyce' function) to warning, otherwise Boyce would say "argument X matches multiple formal arguments" * MilerCalib - added final empty line ============================================ modEvA v3.8.4 (committed 2022-10-28) -> CRAN ============================================ ## Modified functions: * applyThreshold - added "as.numeric" to 'sort(tresh)', otherwise it was character if one of the input 'thresh' criteria was character, which generated errors downstream ========================================== modEvA v3.8.3 (committed 2022-10-27) ========================================== ## Modified functions: * predPlot - 'thresh' can now be set to NA or NULL - legend is also changed accordingly * optiThresh - warning emitted if obs are all 0 or all 1 (which may originate errorr) - propagates to optiPair ## Other modified files: * MillerCalib.Rd - added explanations and examples on applicability to non-logit link models and non-probability predictions ========================================== modEvA v3.82 (committed 2022-07-21) ========================================== ## Modified functions: * varPart - added argument 'cor.method' - changed previous argument name 'method' to 'pred.type', to avoid confusion with the above ========================================== modEvA v3.81 (committed 2022-07-20) ========================================== ## Modified functions: * varPart - argument 'model' now accepts glms of any family (but properly tested only for binomial!) ========================================== modEvA v3.8 (committed 2022-07-20) ========================================== ## Modified functions: * varPart - new arguments 'model', 'groups', 'method', 'return.models' - output table now doesn't say "pure" or "overlap", just the factor names separated by "_" - argument 'coloured' changed to 'colr' to avoid English/American spelling issues ========================================== modEvA v3.7 (committed 2022-07-10) ========================================== ## Modified functions: * applyThreshold - 'thresh' can now be of length 2, for high, intermediate and low predictions ## Other modified files: * DESCRIPTION - added reference about the method, as suggested by Uwe Ligges after previous CRAN submission ========================================== modEvA v3.6 (committed 2022-06-14) ========================================== ## New functions: * confusionMatrix ## Modified functions: * confusionLabel - added 'getThreshold' to include all 'thresh' options - added 'interval' and 'quant' arguments for the above - 'rm.dup' now defaults to FALSE to assume less * threshMeasures - moved part of the code to new 'confusionMatrix' function, which is now called from here ========================================== modEvA v3.5 (committed 2022-05-16) -> CRAN ========================================== ## New functions: * applyThreshold ## Modified functions: * Boyce - warning emitted and points coloured red for bins with less than 30 values * standard01 (benefitting threshMeasures) - added is.finite(score) to avoid error when NAs produced by kappa integer overflow * getThreshold, inputMunch - 'obs' can be null if only 'pred' is needed * predPlot - added call to 'getThreshold' to include more threshold options - added 'cex' argument (default 0.5), plus 'interval' and 'quant' to pass to 'getThreshold' * confusionLabel - output is SpatRaster if input is too (as also implemented in 'applyThreshold') * optiThresh - fixed bug: Precision and Recall were missing from 'goodness.measures' and didn't get optimals ========================================== modEvA v3.4 (committed 2022-05-06) ========================================== ## Modified functions: * AUC, confusionLabel, Dsquared, getBins, HLfit, MillerCalib, optiPair, optiThresh, plotGLM, predDensity, predPlot, RsqGLM, threshMeasures - added arguments 'rm.dup' and 'na.rm' to pass to 'inputMunch' and cascade to 'ptsrast2obspred' * Boyce - renamed arguments 'rm.dupl.classes' and 'rm.dupl.points' to 'rm.dup.classes' and 'rm.dup.points', for better coherence with 'rm.dup' in other functions - added 'na.rm' argument ========================================== modEvA v3.3 (committed 2022-05-03) ========================================== ## New functions: * inputMunch - converts from model or point & raster inputs to obs & pred vectors * getThreshold - computes threshold based on any of a range of published criteria ## Modified functions: * prevalence - added 'unlist' to avoid obscure error when input is one-column tibble instead of vector * confusionLabel, Dsquared, getBins, HLfit, MillerCalib, optiPair, optiThresh, plotGLM, predDensity, predPlot, RsqGLM, threshMeasures - 'obs' and 'pred' can also be presence coordinates and a SpatRaster, respectively * confusionLabel, getBins - added 'na.rm' argument * threshMeasures - 'thresh' argument now calls new 'getThreshold' function to include more options * ptsrast2obspred - checks for duplicates and shows message if there are (if rm.dup = FALSE) - 'rm.dup' now defaults to FALSE, to assume less and to match 'ecospat.boyce' * Boyce - 'rm.dupl.points' now defaults to FALSE, to assume less and to match 'ecospat.boyce' * modEvAmethods - added error message when 'fun' not correctly specified (instead of silent empty result) - help file now provides all available values under 'fun' argument * getModEqn - 'round' replaced with 'signif' to avoid small coeffs becoming zeros (bug report by Jose Carlos Guerrero) * MESS - added 'verbosity' argument ========================================== modEvA v3.2 (committed 2022-03-13) ========================================== ## Modified functions: * AUC - 'obs' and 'pred' can be presence coordinates and a SpatRaster, respectively * Boyce - added argument '...', e.g. for plot 'main' or 'xlim' - argument name 'nclass' replaced with 'n.bins' to match other modEvA functions - argument name 'window.w' replaced with 'bin.width' to accomodate other modEvA functions - default 'nclass = 0' replaced with 'n.bins = NA', and all else accordingly - changed default 'rm.dupl.classes' to FALSE after some checks (e.g. Galpyr RF in SDMB course) * ptsrast2obspred - added argument 'na.rm' and set to TRUE by default - help file now uses 'elev' raster for more clarity ========================================== modEvA v3.1 (committed 2022-02-22) ========================================== ## New functions: - Boyce - ptsrast2obspred ## Modified functions: * AUC - moved 'pred' range check to after NAs are removed to avoid error * predPlot - on exit, restores only 'par(mar)' (the only one changed) rather than all 'par' * RsqGLM - added las = 2 to 'barplot' ========================================== modEvA v3.0 (committed 2021-12-20) -> CRAN ========================================== ## Modified functions: * AUC - warning if 'obs' contains only zeros or only ones (to explain result NaN and no curve on plot) - if the previous occurs, other warnings suppressed (PR curve interpolated past NaN precision, method changed to 'trapezoid', etc.) * MillerCalib - plot text elevated by 0.05 y-axis units, to avoid it being cropped in small plots ## Other changes: - uniformized description of arguments "model", "obs" and "pred" across functions' help files ========================================== modEvA v2.8 (committed 2021-12-03) ========================================== ## Modified functions: * optiPair - message "'pred' must range between 0 and 1" changed from error to warning - added argument 'na.rm' (default TRUE) to avoid NaNs in result MinDiff, MaxSum, etc. - added argument 'exclude.zeros' (default TRUE) to avoid precision-recall "optimal" difference when both 0 or NaN ========================================== modEvA v2.7 (committed 2021-11-26) ========================================== ## New functions: * confusionLabel (to map the confusion matrix) * mod2obspred (to extract 'obs' and 'pred' from model objects) ## Modified functions: * prevalence - message "'event' is not among of the values of 'obs'" changed from error to warning - 'mod2obspred' now used for extracting 'obs' if 'model' is provided * AUC, getBins, HLfit, MillerCalib, optiPair, optiThresh, predDensity, predPlot, threshMeasures - 'model' can now be of class "glm", "gam", "gbm", "randomForest" or "bart" - 'mod2obspred' now used for extracting 'obs' and 'pred' if 'model' is provided ========================================== modEvA v2.6 (committed 2021-11-01) ========================================== ## Modified functions: * predDensity - added 'main' argument for plot title * predPlot - added 'pch' and 'col' arguments for the plot - legend.pos is now "n" by default (legend was ugly and needed box to distinguish from actual plot points) * RsqGLM - added "plot" argument (default TRUE) - added '...' for additional parameters for 'plot' * plotGLM - added 'plot=FALSE' to 'RsqGLM' (used if 'plot.values') * optiThresh - slightly changed plot margins so that bottom axis values are visible ## Other modified files: * www/modEvA-tutorial.html - updated with these recent options ========================================== modEvA v2.5 (committed 2021-10-31) ========================================== ## Modified functions: * AUC - plot.preds=TRUE now plots filled circles with transparency - plot.preds=TRUE now defaults to plotting circles on the curve (rather than both curve and bottom) - size of circles if plot.preds=TRUE changed from 100*prop.preds to 20*sqrt(prop.preds) - x coordinates of circles if plot.preds=TRUE changed from 'thresholds' to 'xx' - if plot.values=TRUE, value is now plotted on the middle bottom for ROC and middle top for PR curve - NaN precision values now coerced to the last non-NaN value (rather than 1), and warning emitted - interpolated (NaN precision) part of the PR curve now plotted besides the curve with actual precision values ========================================== modEvA v2.4 (committed 2021-10-30) ========================================== ## Modified functions: * AUC - fixed wrong PR area when last precision value(s) NaN, by coercing NaN precisions to 1 (thanks to bug report by Ying-Ju Tessa Chen) - deactivated option method = "integrate", which was providing somewhat inaccurate values - added / improved warning messages when 'method' changed to match chosen 'curve' - added warning when method = "trapezoid" and interval >= 0.01, noting result is more accurate with smaller intervals * threshMeasures - added 'ylim' argument with default c(0, 1) to avoid error when no selected measure has finite values ## Other modified files: * inst/CITATION - corrected publication year of 'New measures' paper under 'textVersion' of citation (thanks to bug report by Alba Estrada) ========================================== modEvA v2.3 (committed 2021-09-30) ========================================== ## Modified functions: * prevalence - added 'model' argument, which can now be provided instead of 'obs' - help file includes examples with 'model' and with character 'obs' - error message when 'obs' not a vector and when 'event' not in 'obs' - warning when 'obs' ignored in favour of 'model' ========================================== modEvA v2.2 (committed 2021-09-08) ========================================== ## Modified functions: * MillerCalib - slope value now plotted before (above) intercept (if plot.values = TRUE) * varPart - fixed right margin in 2-factor plot and added option for colouring the circles, following suggestions by Oswald van Ginkel * MESS - bug fix e-mailed by Huijie Qiao ========================================== modEvA v2.1 (committed 2021-01-15) ========================================== ## New functions: * predPlot ## Modified functions: * predDensity - like 'predPlot', 'legend.pos' now allows NA (for no legend) instead of throwing obscure error * optiThresh - added warning when any of 'measures' or 'optimize' not within implemented values ## Other modified files: * MESS.Rd - removed mention that we will implement ExDet for dataframes; added 'ecospat.climan' to 'See also' ## Other changes: * minor corrections to some 'man' files ========================================== modEvA v2.0 (committed 2020-01-21) -> CRAN ========================================== ## 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