Changes in version 7.0-1 (2024-12-10) o CRAN mandated update. o caret will be 20 years old in March of 2026. The package is currently in maintenance mode; the author will fix bugs and make CRAN releases as needed, but there will not be any major features in the package. It will stay on CRAN long-term; it's not going away. Changes in version 6.0-94 (2023-03-21) o Bug fix in how some S3 signatures were designed (for R-devel). o Adrián Panella fixed a bug with `method = "svmRadial" that occured when the SVM probability model failed (issue 1327) o Theodore Pak fixed a bug in glmnet prediction when sparse matrices are used (issue 1315) o Carlos Cinelli fixed two bugs (one for GAMs and another for ranger models) where an error would occur when using trainControl(method = "none") (issue 1307) (issue 1308) Changes in version 6.0-93 (2022-08-09) o CRAN required changes for legacy S typedef Sint. o Xabriel J. Collazo Mojica added check for class probabilities when PR curves are requested (issue 1274). Changes in version 6.0-92 (2022-04-19) o Small maintenance release with required C changes for R 4.2. Changes in version 6.0-91 (2022-03-11) o Small maintenance release with required changes for R 4.2. Changes in version 6.0-90 (2021-10-09) o A ptype element was added to train objects that records the trianing set predictor columns and their types using a zero-row slice. o Updated the internal object containing the subsampling information. The old package dependencies were being used. Changes in version 6.0-89 (2021-09-28) o SMOTE subsampling is now computed via the themis package (issue 1226). o For SA feature selection, a better warning is given when there are too few iterations for computing differences (issue 1247). o Better error message when pacakges are missing (issue 1246). o e1071 was promoted from Suggests to Imports (issue 1238). Changes in version 6.0-88 (2021-05-15) o Fixed cases where the "corr" filter was not run in preProcess(). o Prediction type bug for Poisson glm model was fixed (issue 1231). o Fixed a PreProcess() bug related to a single PCA component (issue 1181). o Fixed random forest bugs related to rfe() (issue #1077), (issue 1062). o RuleFit was added via the pre package (issue 1218). o Bugs were fixed where MAE was not treated as a minimization metric (issue 1224). Changes in version 6.0-87 o The ordinalNet model was given an additional tuning parameter modeltype. Changes in version 6.0-86 (2020-03-20) o Small release for stringsAsFactors = TRUE in R-4.0 Changes in version 6.0-85 (2020-01-07) o Internal changes required by r-devel for new matrix class structure. o Michael Mayer contributed a faster version of groupKFold(). (issue 1108) o A typo in a variable name was fixed. (issue 1087) o Removed some warnings related to contrasts. o Temporarily moved ROC calculations back to the pROC package related to JackStat/ModelMetrics#30. (issue 1105) Changes in version 6.0-84 (2019-04-27) o Another new version was related to character encodings. Changes in version 6.0-83 (2019-04-18) o A new version was requested by CRAN since en dashes were used in the documentation. o A bug was fixed where, for some recipes that involve class imbalance sampling, the resampling indicies were computed incorrectly (issue 1030). o train now removes duplicate models in the tuning grid. Duplicates could occur for models with discrete parameters. Changes in version 6.0-82 (2019-03-26) o Immediate and required updates related to the different behavior of sample in R >= 3.6. o sbf, gafs, and safs now accept recipes as inputs. A few sections of documentation were added to the bookdown site o A bug was fixed in simulated annealing feature selection where the number of variables perturbed was based on the total number of variables instead of the more appropriate number of variables in the current subset. o Convenience functions ggplot.gafs and ggplot.safs were added. o learning_curve_dat now has a real name. o The earth and ctree models were udpdate to fix bugs (issue 1022)(issue 1018). o When a model has the same value for its resamples, plot.resamples and ggplot.resamples now produce an estimate and missing values for the intervals (instead of failing) (issue 1007) Changes in version 6.0-81 (2018-11-20) o The blackboost code gained a dependency on partykit due to changes in mboost. o The internals were updated to work with the latest version of the recipes package. o Jason Muhlenkamp added better error messages for misspecified tuning parameters (issue 956). o Two bugs in random forests with RFE were fixed in (issue 942). o When correlation filters are used in preProcess, constant (i.e. zero-variance) columns are first removed to avoid errors (issue 966). o A bug was fixed when train models using weights were updated (issue 935). o Benjamin Allévius added more statistics to the output of thresholder (issue 938). o A bug was fixed that occurred when indexFinal was used where the recipe that was saved was created using the entire training set (issue 928). Changes in version 6.0-80 (2018-05-26) o Two bugs associated with varImp in lm (issue 858) and in bartMachine were fixed by hadjipantelis. o SMOTE sampling now works with tibbles (issue 875) o rpart now a dependency due to new CRAN policies. o Added a ggplot method for resamples that produces confidence intervals. o hadjipantelis added some fixes for mxnet models (issue 887). Changes in version 6.0-79 (2018-03-29) o keras and mxnet models have better initialization of parameters pr 765 o The range preprocessing method can scale the data to an arbitrary range. Thanks to Sergey Korop. o The spatial sign transformation will now operation on all non-missing predictors. Thanks to Markus Peter Auer (issue 789). o A variety of small changes were made to work with the new version of the gam package (issue 828). o The package vignette was changed to HTML. o A big was fixed for computing variable importance scores with the various PLS methods (issue 848). o Fixed a drop = FALSE bug that occurred when computing class probabilities (issue 849). o An issue with predicting probabilities with multinom and one observation was fixed (issue 827). o A bug in the threshold calculation for choosing the number of PCA components was resolved (issue 825). o Models mlpML and mlpWeightDecayML now ignore layers with zero units. For example, if the number of layers was specified to be c(5, 0, 3) a warning is issued and the architecture is converted to c(5, 3) (issue 829). o svmLinearWeights2 and svmLinear3 may have chosen the incorrect SVM loss function. This was found by Dirk Neumann (issue 826) o bnclassify models awtan and awnb were updated since they previously used deprecated functions. All bnclassify models now require version 0.3.3 of that package or greater (issue 815). o confusionMatrix.default not requires data and reference to be factors and will throw an error otherwise. Previously, the vectors were converted to factors but this resulted in too many bug reports and misuse. o xyplot.resample did not pass the dots to the underlying plot function (issue 853). o A bug with model xgbDART was fixed by hadjipantelis. Changes in version 6.0-78 (2017-12-10) o A number of changes were made to the underlying model code to repair problems caused by the previous version. In essence, unless the modeling package was formally loaded, the model code would fail in some cases. In the vast majority of cases, train will not load the package (but will load the namespace). There are some exceptions where this is not possible, including bam, earth, gam, gamLoess, gamSpline, logicBag, ORFlog, ORFpls, ORFridge, ORFsvm, plsRglm, RSimca, rrlda, spikeslab, and others. These are noted in ?models and in the model code itself. The regression tests now catch these issues. o The option to control the minimum node size to models ranger and Rborist was added by hadjipantelis (issue 732). o The rule-based model GFS.GCCL was removed from the model library. o A bug was fixed affecting models using the sparsediscrim package (i.e. dda and rlda) where the class probability values were reversed. (issue 761). o The keras models now clear the session prior to each model fit to avoid problems. Also, on the last fit, the model is serialized so that it can be used between sessions. The predict code will automatically undo this encoding so that the user does not have to manually intervene. o A bug in twoClassSummary was fixed that prevents failure when the class level includes "y" (issue 770). o The preProcess function can now scale variables to a range where the user can set the high and low values (issue 730). Thanks to Sergey Korop. o Erwan Le Pennec fixed some issues when train was run using some parallel processing backends (e.g. doFuture and doAzureParallel) (issue 748). o Waleed Muhanna found and fixed a bug in twoClassSim when irrelevant variables were generated. (issue 744). o hadjipantelis added the DART model (aka "Dropouts meet Multiple Additive Regression Trees") with the model code xgbDART (issue 742). o Vadim Khotilovich updated predict.dummyVars to run faster with large datasets with many factors (issue 727). o spatialSign now has the option of removing missing data prior to computing the norm (issue 789). o The various earth models have been updated to work with recent versions of that package, including multi-class glm models (issue 779). Changes in version 6.0-77 (2017-09-07) o Two neural network models (containing up to three hidden layers) using mxnet were added; mxnet (optimiser: SGD) and mxnetAdam (optimiser: ADAM). o A new method was added for train so that recipes can be used to specify the model terms and preprocessing. Alexis Sardá provided a great deal of help converting the bootstrap optimism code to the new workflows. A new chapter was added to the package website related to recipes. o The Yeo-Johnson transformation parameter estimation code was rewritten and not longer requires the car package. o The leave-one-out cross-validation workflow for train has been harmonized with the other resampling methods in terms of fault tolerance and prediction trimming. o train now uses different random numbers to make resamples. Previously, setting the seed prior to calling train should result in getting the same resamples. However, if train loaded or imported a namespace from another package, and that startup process used random numbers, it could lead to different random numbers being used. See (issue 452) for details. Now, train gets a separate (and more reproducible) seed that will be used to generate the resamples. However, this may effect random number reproducibility between this version and previous versions. Otherwise, this change should increase the reproducibility of results. o Erwan Le Pennec conducted the herculean task of modifying all of the model code to call by namespace (instead of fully loading each required package). This should reduce naming conflicts (issue 701). o MAE was added as output metric for regression tasks through postResample and defaultSummary by hadjipantelis. The function is now exposed to the users. (issue 657). o More average precision/recall statistics were added to multiClassSummary (issue 697). o The package website code was updated to use version 4 of the D3 JS library and now uses heatmaply to make the interactive heatmap. o Added a ggplot method for lift objects (and fixed a bug in the lattice version of the code) for (issue 656). o Vadim Khotilovich made a change to speed up predict.dummyVars (issue 727). o The model code for ordinalNet was updated for recent changes to that package. o oblique.tree was removed from the model library. o The default grid generation for rotation forest models now provides better values of K. o The parameter ranges for AdaBag and AdaBoost.M1 were changed; the number of iterations in the default grids have been lowered. o Switched to non-formula interface in ranger. Also, another tuning parameter was added to ranger (splitrule) that can be used to change the splitting procedure and includes extremely randomized trees. This requires version 0.8.0 of the ranger package. (issue 581) o A simple "null model" was added. For classification, it predictors using the most prevalent level and, for regression, fits an intercept only model. (issue 694) o A function thresholder was added to analyze the resample results for two class problems to choose an appropriate probability cutoff a la (issue 224). o Two neural network models (containing a single hidden layers) using tensorflow/keras were added. mlpKerasDecay uses standard weight decay while mlpKerasDropout uses dropout for regularization. Both use RMSProp optimizer and have a lot of tuning parameters. Two additional models, mlpKerasDecayCost and mlpKerasDropoutCost, are classification only and perform cost-sensitive learning. Note that these models will not run in parallel using caret's parallelism and also will not give reproducible results from run-to-run (see ). o The range for one parameter (gamma) was modified in the mlpSGD model code. o A bug in classification models with all missing predictions was fixed (found by andzandz11). (issue 684) o A bug preventing preprocessing to work properly when the preprocessing transformations are related to individual columns only fixed by Mateusz Kobos in (issue 679). o A prediction bug in glm.nb that was found by jpclemens0 was fixed (issue 688). o A bug was fixed in Self-Organizing Maps via xyf for regression models. o A bug was fixed in rpartCost related to how the tuning parameter grid was processed. o A bug in negative-binomial GLM models (found by jpclemens0) was fixed (issue 688). o In trainControl, if repeats is used on methods other than "repeatedcv" or "adaptive_cv", a warning is issued. Also, for method other than these two, a new default (NA) is given to repeats. (issue 720). o rfFuncs now computes importance on the first and last model fit. (issue 723) Changes in version 6.0-76 (2017-04-18) o Monotone multi-layer perceptron neural network models from the monmlp package were added (issue 489). o A new resampling function (groupKFold) was added (issue 540). o The bootstrap optimism estimate was added by Alexis Sarda (issue 544). o Bugs in glm, glm.nb, and lm variable importance methods that occur when a single variable is in the model (issue 543). o A bug in filterVarImp was fixed where the ROC curve AUC could be much less than 0.50 because the directionality of the predictor was not taken into account. This will artificially increase the importance of some non-informative predictors. However, the bug might report the AUC for an important predictor to be 0.20 instead of 0.80 (issue 565). o multiClassSummary now reports the average F score (issue 566). o The RMSE and R2 are now (re)exposed to the users (issue 563). o A caret bug was discovered by Jiebiao Wang where glmboost, gamboost, and blackboost models incorrectly reported the class probabilities (issue 560). o Training data weights support was added to xgbTree model by schistyakov. o Regularized logistic regression through Liblinear (LiblineaR::LiblineaR) using L1 or L2 regularization were added by hadjipantelis. o A bug related to the ordering of axes labels in the heatmap plot of training results was fixed by Mateusz Dziedzic in (issue 620). o A variable importance method for model averaged neural networks was added. o More logic was added so that the predict method behaves well when a variable is subtracted from a model formula from (issue 574). o More documentation was added for the class2ind function ((issue 592)). o Fixed the formatting of the design matrices in the dummyVars man file. o A note was added to ?trainControl about using custom resampling methods ((issue 584)). o A bug was fixed related to SMOTE and ROSE sampling with one predictor ((issue 612)). o Due to changes in the kohonen package, the bdk model is no longer available and the code behind the xyf model has changes substantially (including the tuning parameters). Also, when using xyf, a check is conducted to make sure that a recent version of the kohonen package is being used. o Changes to xgbTree and xgbLinear to help with sparse matrix inputs for (issue 593). Sparse matrices are not allowed when preprocessing or subsampling are used. o Several PLS models were using the classical orthogonal scores algorithm when discriminant analysis was conducted (despite using simpls, widekernelpls, or kernelpls). Now, the PLSDA model estimation method is consistent with the method requested ((issue 610)). o Added Multi-Step Adaptive MCP-Net (method = "msaenet") for (issue 561). o The variable importance score for linear regression was modified so that missing values in the coefficients are converted to zero. o In train, x is now required to have column names. Changes in version 6.0-73 (2016-11-10) o Negative binomial generalized linear models (MASS:::glm.nb) were added (issue 476) o mnLogLoss now returns a named vector ((issue 514), bug found by Jay Qi) o A bunch of method/class related bugs induced by the previous version were fixed. Changes in version 6.0-72 (2016-11-01) o The inverse hyperbolic sine transformation was added to preProcess (issue 56) o Tyler Hunt moved the ROC code from the pROC package to the ModelMetrics package which should make the computations more efficient (issue 482). o train does a better job of respecting the original format of the input data (issue 474) o A bug in bdk and xyf models was fixed where the appropriate number of parameter combinations are tested during random search. o A bug in rfe was fixed related to neural networks found by david-machinelearning (issue 485) o Neural networks via stochastic gradient descent (method = "mlpSGD") was adapted for classification and a variable importance calculation was added. o h2o versions of glmnet and gradient boosting machines were added with methods "glmnet\_h2o" and "gbm\_h2o". These methods are not currently optimized. (issue 283) o The fuzzy rule-based models (WM, SLAVE, SBC, HYFIS, GFS.THRIFT, GFS.LT.RS, GFS.GCCL, GFS.FR.MOGUL, FS.HGD, FRBCS.W, FRBCS.CHI, FIR.DM, FH.GBML, DENFIS, and ANFIS) were modified so that the user can pass in the predictor ranges using the range.data argument to those functions. (issue 498) o A variable importance method was added for boosted generalized linear models (issue 493) o preProcess now has an option to filter out highly correlated predictors. o trainControl now has additional options to modify the parameters of near-zero variance and correlation filters. See the preProcOptions argument. o The rotationForest and rotationForestCp methods were revised to evaluate only _feasible_ values of the parameter K (the number of variable subsets). The underlying rotationForest function reduces this parameter until values of K divides evenly into the number of parameters. o The skip option from createTimeSlices was added to trainControl (issue 491) o xgb.train's option subsample was added to the xgbTree model (issue 464) Changes in version 6.0-71 (2016-08-05) o Precision, recall, and F measure functions were added along with one called prSummary that is analogous to twoClassSummary. Also, confusionMatrix gains an argument called mode that dictates what output is shown. o schistyakov added additional tuning parameters to the robust linear model code (issue 454). Also for rlm and lm schistyakov added the ability to tune over the intercept/no intercept model. o Generalized additive models for very large datasets (bam in mgcv) was added (issue 453) o Two more linear SVM models were added from the LiblineaR package with model codes svmLinear3 and svmLinearWeights2 ((issue 441)) o The tau parameter was added to all of the least square SVM models ((issue 415)) o A new data set (called scat) on animal droppings was added. o A significant bug was fixed where the internals of how R creates a model matrix was ignoring na.action when the default was set to na.fail (issue 461). This means that train will now immediately fail if there are any missing data. To use imputation, use na.action = na.pass and the imputation method of your choice in the preProcess argument. Also, a warning is issued if the user asks for imputation but uses the formula method and excludes missing data in na.action Changes in version 6.0-70 (2016-06-13) o Based on a comment by Alexis Sarda, method = "ctree2" does not fix mincriterion = 0 and tunes over this parameter. For a fixed depth, mincriterion can further prune the tree (issue 409). o A bug in KNN imputation was fixed (found by saviola777) that occurred when a factor predictor was in the data set (issue 404). o Infrastructure changes were made so that train tries harder to respect the original class of the outcome. For example, if an ordered factor is used as the outcome with a modeling function that treats is as an unordered factor, the model still produces an ordered factor during prediction. o The ranger code now allows for case weights (issue 414). o twoClassSim now has an option to compute ordered factors. o High-dimensional regularized discriminant analysis and, regularized linear discriminant analysis, and several variants of diagonal discriminant analysis from the sparsediscrim package were added (method = "hdrda", method = "rlda", and method = "dda", respectively) (issue 313). o A neural network regression model optimized by stochastic gradient decent from the FCNN4R package was added. The model code is mlpSGD. o Several models for ordinal outcomes were added: rpartScore (from the rpartScore package), ordinalNet (ordinalNet), vglmAdjCat (VGAM), vglmContRatio (VGAM), and vglmCumulative (VGAM). Note that, for models that load VGAM, there is a conflict such that the predictors class code from caret is masked. To use that method, you can use caret:::predictors.train() instead of predictors(). o Another high performance random forest package (Rborist) was exposed through caret. The model code is method = "Rborist" (issue 418) o Xavier Robin fixed a bug related to the area under the ROC curve in (issue 431). o A bug in print.train was fixed when LOO CV was used (issue 435) o With RFE, a better error message drafted by mikekaminsky is printed when the number of importance measures is off (issue 424) o Another bug was fixed in estimating the prediction time when the formula method was used (issue 420). o A linear SVM model was added that uses class weights. o The linear SVM model using the e1071 package (method = "svmLinear2") had the gamma parameter for the RBF kernel removed. o Xavier Robin committed changes to make sure that the area under the ROC is accurately estimated (issue 431) Changes in version 6.0-68 (2016-04-11) o print.train no longer shows the standard deviation of the resampled values unless the new option is used (print.train(, showSD = TRUE)). When shown, they are within parentheses (e.g. "4.24 (0.493)"). o An adjustment the innards of adaptive resampling was changed so that the test for linear dependencies is more stringent. o A bug in the bootstrap 632 estimate was found and fixed by Alexis Sarda (issue 349) (issue 353). o The cforest module's oob element was modified based on another bug found by Alexis Sarda (issue 351). o The methods for bagEarth, bagEarthGCV, bagFDA, bagFDAGCV, earth, fda, and gcvEarth models have been updates so that case-weights can be used. o The rda module contained a bug found by Eric Czech (issue 369). o A bug was fixed for printing out the resampling details with LGOCV found by github user zsharpm (issue 366) o A new data set was added (data(Sacramento)) with sale prices of homes. o Another adaboost algorithm (method = "adaboost" from the fastAdaboost package) was added (issue 284). o Yet another boosting algorithm (method = "deepboost" from the deepboost package) was added (issue 388). o Alexis Sarda made changes to the confusion matrix code for train, rfe, and sbf objects that more rationally normalizes the resampled tables (issue 355). o A bug in how RSNNS perceptron models were tuned (found by github user smlek) was fixed (issue 392). o A bug in computing the bootstrap 632 estimate was fixed (found by Stu) (issue 382). o John Johnson contributed an update to xgbLinear (issue 372). o Resampled confusion matrices are not automatically computed when there are 50 or more classes due to the storage requirements ((issue 356)). However, the relevant functions have been updated to use the out-of-sample predictions instead (when the user asks for them to be returned by the function). o Some changes were made to predict.train to error trap (and fix) cases when predictions are requested without referencing a newdata object (issue 347). o Github user pverspeelt identified a bug in our model code for glmboost (and gamboost) related to the mstop function modifying the model object in memory. It was fixed (issue 396). o For (issue 346), an option to select which samples are used to fit the final model, called indexFinal, was added to trainControl. o For issue (issue 390) found by JanLauGe, a bug was fixed in dummyVars related to the names of the resulting data set. o Models rknn and rknnBel were removed since their package is no longer on CRAN. Changes in version 6.0-66 o Model averaged naive Bayes (method = "manb") from the bnclassify package was added. o blackboost was updated to work with outcomes with 3+ classes. o A new model rpart1SE was added. This has no tuning parameters and resamples the internal rpart procdure of pruning using the one standard error method. o Another model (svmRadialSigma) tunes over the cost parameter and the RBF kernel parameter sigma. In the latter case, using tuneLength will, at most, evaluate six values of the kernel parameter. This enables a broad search over the cost parameter and a relatively narrow search over sigma. o Additional model tags for "Accepts Case Weights", "Two Class Only", "Handle Missing Predictor Data", "Categorical Predictors Only", and "Binary Predictors Only" were added. In some cases, a new model element called "notes" was added to the model code. o A pre-processing method called "conditionalX" was added that eliminates predictors where the conditional distribution (X|Y) for that predictor has a single value. See the checkConditionalX function for details. This is only used for classification. (issue 334) o A bug in the naive Bayes prediction code was found by github user pverspeelt and was fixed. (issue 345) o Josh Brady (doublej2) found and fixed an issue with DummyVars (issue 344) o A bug related to recent changes to the ranger package was fixed (issue 320) o Dependencies on external software can now be checked in the model code. See pythonKnnReg for an example. This also removes the overall package dependency on rPython (issue 328). o The tuning parameter grid for enpls and enpls.fs were changed to avoid errors. o A bug was fixed (issue 342) where the data used for prediction was inappropriately converted from its original class. o Matt (aka washcycle) added option to return column names to nearZeroVar function o Homer Strong fixed varImp for glmnet models so that they return the absolute value of the regression coefficients (issue 173) (issue 190) o The basic naive Bayes method (method = "nb") gained a tuning parameter, adjust, that adjusts the bandwidth (see ?density). The parameter is ignored when usekernel = FALSE. Changes in version 6.0-62 (2015-11-23) o From the randomGLM package, a model of the same name was added. o From monomvn package, models for the Bayesian lasso and ridge regression were added. In the latter case, two methods were added. blasso creates predictions using the mean of the posterior distributions but sets some parameters specifically to zero based on the tuning parameter called sparsity. For example, when sparsity = .5, only coefficients where at least half the posterior estimates are nonzero are used. The other model, blassoAveraged, makes predictions across all of the realizations in the posterior distribution without coercing any coefficients to zero. This is more consistent with Bayesian model averaging, but is unlikely to produce very sparse solutions. o From the spikeslab package, a regression model was added that emulates the procedure used by cv.spikeslab where the tuning variable is the number of retained predictors. o A bug was fixed in adaptive resampling (found by github user elephann) (issue 304) o Fixed another adaptive resampling bug flagged by github user elephann related to the latest version of the BradleyTerry2 package. Thanks to Heather Turner for the fix (issue 310) o Yuan (Terry) Tang added more tuning parameters to xgbTree models. o Model svmRadialWeights was updated to allow for class probabilities. Previously, kernlab did not change the probability estimates when weights were used. o A ggplot2 method for varImp.train was added (issue 231) o Changes were made for the package to work with the next version of ggplot2 (issue 317) o Github user fjeze added new models mlpML and mlpWeightDecayML that extend the existing RSNNS models to multiple layers. fjeze also added the gamma parameter to the svmLinear2 model. o A function for generating data for learning curves was added. o The range of SVM cost values explored in random search was expanded. Changes in version 6.0-58 (2015-10-22) o A major bug was fixed (found by Harlan Harris) where pre-processing objects created from versions of the package prior to 6.0-57 can give incorrect results when run with 6.0-57 (issue 282). o preProcess can now remove predictors using zero- and near zero-variance filters via (method values of "zv" and "nzv"). When used, these filters are applied to numeric predictors prior to all other pre-processing operations. o train now throws an error for classification tasks where the outcome has a factor level with no observed data (issue 260). o Character outcomes passed to train are not converted to factors. o A bug was found and fixed in this package's class probability code for gbm models when a single multinomial observation is predicted (issue 274). o A new option to ggplot.train was added that highlights the optimal tuning parameter setting in the cases where grid search is used (thanks to Balaji Iyengar (github: bdanalytics)). o In trainControl, the argument savePredictions can now be character values ("final", "all" or "none"). Logicals can still be used and match to "all" or "none". Changes in version 6.0-57 (2015-10-11) o Hyperparameter optimization via random search is now availible. See the new help page for examples and syntax. o preProcess now allows (but ignores) non-numeric predictor columns. o Models were added for optimal weighted and stabilized nearest neighbor classifiers from the snn package were added with model codes snn and ownn o Random forests using the excellent ranger package were added (method = "ranger") o An additional variation of rotation forests was added (rotationForest2) that also tunes over cp. Unfortunately, the sub-model trick can't be utilized in this instance. o Kernelized distance weighted discriminant analysis models from kerndwd where added (dwdLieanr, dwdPoly, and dwdRadial) o A bug was fixed with rfe when train was used to generate a classification model but class probabilities were not (or could not be) generated (issue 234). o Can Candan added a python model sklearn.neighbors.KNeighborsRegressor that can be accessed via train using the rPython package. The python modules sklearn and pandas are required for this to run. o Jason Aizkalns fixed a bunch of typos. o MarwaNabil found a bug with lift and missing values (issue 225). This was fixed such that missing values are removed prior to the calculations (within each model) o Additional options were added to LPH07_1 so that two class data can also be simulated and predictors are converted to factors. o The model-specific code for computing out-of-bag performance estimates were moved into the model code library (issue 230). o A variety of naive Bayes and tree augmented naive Bayes classifier from the bnclassify package were added. Variations include simple models (methods labeled as "nbDiscrete" and "tan"), models using attribute weighting ("awnb" and "awtan"), and wrappers that use search methods to optimize the network structure ("nbSearch" and "tanSearch"). In each case, the predictors and outcomes must all be factor variables; for that reason, using the non-formula interface to train (e.g. train(x, y)) is critical to preserve the factor structure of the data. o A function called multiClassSummary was added to compute performance values for problems with three or more classes. It works with or without predicted class probabilities (issue 107). o confusionMatrix was modified to deal with name collisions between this package and RSNNS (issue 256). o A bug in how the LVQ tune grid is filtered was fixed. o A bug in preProcess for ICA and PCA was fixed. o Bugs in avNNet and pcaNNet when predicting class probabilities were fixed (issue #261). Changes in version 6.0-52 (2015-07-17) o A new model using the randomForest and inTrees packages called rfRules was added. A basic random forest model is used and then is decomposed into rules (of user-specified complexity). The inTrees package is used to prune and optimize the rules. Thanks to Mirjam Jenny who suggested the workflow. o Other new models (and their packages): bartMachine (bartMachine), rotationForest (rotationForest), sdwd (sdwd), loclda (klaR), nnls (nnls), svmLinear2 (e1071), rqnc (rqPen), and rqlasso (rqPen) o When specifying your own resampling indices, a value of method = "custom" can be used with trainControl for better printing. o Tim Lucas fixed a bug in avNNet when bag = TRUE o Fixed a bug found by ruggerorossi in method = "dnn" with classification. o A new option called sampling was added to trainControl that allows users to subsample their data in the case of a class imbalance. Another help page was added to explain the features. o Class probabilities can be computed for extraTrees models now. o When PCA pre-processing is conducted, the variance trace is saved in an object called trace. o More error traps were added for common mistakes (e.g. bad factor levels in classification). o An internal function (class2ind) that can be used to make dummy variables for a single factor vector is now documented and exported. o A bug was fixed in the xyplot.lift where the reference line was incorrectly computed. Thanks to Einat Sitbon for finding this. o A bug related to calculating the Box-Cox transformation found by John Johnson was fixed. o github user EdwinTh developed a faster version of findCorrelation and found a bug in the original code. findCorrelation has two new arguments, one of which is called exact which defaults to use the original (fixed) function. Using exact = FALSE uses the faster version. The fixed version of the "exact" code is, on average, 26-fold slower than the current version (for 250x250 matrices) although the average time for matrices of this size was only 26s. The exact version yields subsets that are, one average, 2.4 percent smaller than the other versions. This difference will be more significant for smaller matrices. The faster ("approximate") version of the code is 8-fold faster than the current version. o github user slyuee found a bug in the gam model fitting code. o Chris Kennedy fixed a bug in the bartMachine variable importance code. Changes in version 6.0-47 (2015-05-06) o CHAID from the R-Forge package CHAID o Models xgbTree amd xgbLinear from the xgboost package were added. That package is not on CRAN and can be installed from github using the devtools package and install_github('dmlc/xgboost',subdir='R-package'). o dratewka enabled rbf models for regression. o A summary function for the multinomial likelihood called mnLogLoss was added. o The total object size for preProces objects that used bagged imputation was reduced almost 5-fold. o A new option to trainControl called trim was added where, if implemented, will reduce the model's footprint. However, features beyond simple prediction may not work. o A rarely occurring bug in gbm model code was fixed (thanks to Wade Cooper) o splom.resamples now respects the models argument o A new argument to lift called cuts was added to allow more control over what thresholds are used to calculate the curve. o The cuts argument of calibration now accepts a vector of cut points. o Jason Schadewald noticed and fixed a bug in the man page for dummyVars o Call objects were removed from the following models: avNNet, bagFDA, icr, knn3, knnreg, pcaNNet, and plsda. o An argument was added to createTimeSlices to thin the number of resamples o The RFE-related functions lrFuncs, lmFuncs, and gamFuncs were updated so that rfe accepts a matrix x argument. o Using the default grid generation with train and glmnet, an initial glmnet fit is created with alpha = 0.50 to define the lambda values. o train models for "gbm", "gam", "gamSpline", and "gamLoess" now allow their respective arguments for the outcome probability distribution to be passed to the underlying function. o A bug in print.varImp.train was fixed. o train now returns an additional column called rowIndex that is exposed when calling the summary function during resampling. o The ability to compute class probabilities was removed from the rpartCost model since they are unlikely to agree with the class predictions. o extractProb no longer redundantly calls extractPrediction to generate the class predictions. o A new function called var_seq was added that finds a sequence of integers that can be useful for some tuning parameters such as random forests mtry. Model modules were update to use the new function. o n.minobsinnode was added as a tuning parameter to gbm models. o For models using out-of-bag resampling, train now properly checks the metric argument against the names of the measured outcomes. o Both createDataParition and createFolds were modified to better handle cases where one or more class have very low numbers of data points. Changes in version 6.0-41 (2015-01-03) o The license was changed to GPL (>= 2) to accommodate new code from the GA package. o New feature selection functions gafs and safs were added, along with helper functions and objects, were added. The package HTML was updated to expand more about feature selection. o From the adabag package, two new models were added: AdaBag and AdaBoost.M1. o Weighted subspace random forests from the wsrf package was added. o Additional bagged FDA and MARS models were added (model codes bagFDAGCV and bagEarthGCV) were added that use the GCV statistic to prune the model. This leads to memory reductions during training. o The model code for ada had a bug fix applied and the code was adapted to use the "sub-model trick" so it should train faster. o A bug was fixed related to imputation when the formula method is used with train o The old drop = FALSE bug was fixed in getTrainPerf o A bug was fixed for custom models with no labels. o A bug fix was made for bagged MARS models when predicting probabilities. o In train, the argument last was being incorrectly set for the last model. o Reynald Lescarbeau refactored findCorrelation to make it faster. o The apparent performance values are not reported by print.train when the bootstrap 632 estimate is used. o When a required package is missing, the code stops earlier with a more explicit error message. Changes in version 6.0-37 (2014-11-10) o Brenton Kenkel added ordered logistic or probit regression to train using method = "polr" from MASS o LPH07_1 now encodes the noise variables as binary o Both rfe and sbf get arguments for indexOut for their control functions. o A reworked version of nearZerVar based on code from Michael Benesty was added the old version is now called nzv that uses less memory and can be used in parallel. o The adaptive mixture discriminant model from the adaptDA package was added as well as a robust mixture discriminant model from the robustDA package. o The multi-class discriminant model using binary predictors in the binda package was added. o Ensembles of partial least squares models (via the enpls) package was added. o A bug using gbm with Poisson data was fixed (thanks to user eriklampa) o sbfControl now has a multivariate option where all the predictors are exposed to the scoring function at once. o A function compare_models was added that is a simple comparison of models via diff.resamples). o The row names for the variables component of rfe objects were simplified. o Philipp Bergmeir found a bug that was fixed where bag would not run in parallel. o predictionBounds was not implemented during resampling. Changes in version 6.0-35 (2014-08-24) o A few bug fixes to preProcess were made related to KNN imputation. o The parameter labels for polynomial SVM models were fixed o The tags for dnn models were fixed. o The following functions were removed from the package: generateExprVal.method.trimMean, normalize.AffyBatch.normalize2Reference, normalize2Reference, and PLS. The original code and the man files can be found at https://github.com/topepo/caret/tree/master/deprecated. o A number of changes to comply with section 1.1.3.1 of "Writing R Extensions" were made. Changes in version 6.0-34 (2014-08-15) o For the input data x to train, we now respect the class of the input value to accommodate other data types (such as sparse matrices). There are some complications though; for pre-processing we throw a warning if the data are not simple matrices or data frames since there is some infrastructure that does not exist for other classes( e.g. complete.cases). We also throw a warning if returnData <- TRUE and it cannot be converted to a data frame. This allows the use of sparse matrices and text corpus to be used as inputs into that function. o plsRglm was added. o From the frbs, the following rule-based models were added: ANFIS, DENFIS, FH.GBML, FIR.DM, FRBCS.CHI, FRBCS.W, FS.HGD, GFS.FR.MOGAL, GFS.GCCL, GFS.LTS, GFS.THRIFT, HYFIS, SBC and WM. Thanks to Lala Riza for suggesting these and facilitating their addition to the package. o From the kernlab package, SVM models using string kernels were added: svmBoundrangeString, svmExpoString, svmSpectrumString o A function update.rfe was added. o cluster.resamples was added to the namespace. o An option to choose the metric was added to summary.resamples. o prcomp.resamples now passed ... to prcomp. Also the call to prcomp uses the formula method so that na.action can be used. o The function resamples was enhanced so that train and rfe models that used returnResamp="all" subsets the resamples to get the appropriate values and issues a warning. The function also fills in missing model names if one or more are not given. o Several regression simulation functions were added: SLC14_1, SLC14_2, LPH07_1 and LPH07_2 o print.train was re-factored so that format.data.frame is now used. This should behave better when using knitr. o The error message in train.formula was improved to provide more helpful feedback in cases where there is at least one missing value in each row of the data set. o ggplot.train was modified so that groups are distinguished by color and shape. o Options were added to plot.train and ggplot.train called nameInStrip that will print the name and value of any tuning parameters shown in panels. o A bug was fixed by Jia Xu within the knn imputation code used by preProcess. Changes in version 6.0-30 (2014-06-04) o A missing piece of documentation in trainControl for adaptive models was filled in. o A warning was added to plot.train and ggplot.train to note that the relationship between the resampled performance measures and the tuning parameters can be deceiving when using adaptive resampling. o A check was added to trainControl to make sure that a value of min makes sense when using adaptive resampling. Changes in version 6.0-29 (2014-05-29) o A man page with the list of models available via train was added back into the package. See ?models. o Thoralf Mildenberger found and fixed a bug in the variable importance calculation for neural network models. o The output of varImp for pamr models was updated to clarify the ordering of the importance scores. o getModelInfo was updated to generate a more informative error message if the user looks for a model that is not in the package's model library. o A bug was fixed related to how seeds were set inside of train. o The model "parRF" (parallel random forest) was added back into the library. o When case weights are specified in train, the hold-out weights are exposed when computing the summary function. o A check was made to convert a data.table given to train to a data frame (see ). Changes in version 6.0-25 o Changes were made that stopped execution of train if there are no rows in the data (changes suggested by Andrew Ziem) o Andrew Ziem also helped improve the documentation. Changes in version 6.0-24 (2014-02-16) o Several models were updated to work with case weights. o A bug in rfe was found where the largest subset size have the same results as the full model. Thanks to Jose Seoane for reporting the bug. Changes in version 6.0-22 (2014-01-18) o For some parallel processing technologies, the package now export more internal functions. o A bug was fixed in rfe that occurred when LOO CV was used. o Another bug was fixed that occurred for some models when tuneGrid contained only a single model. Changes in version 6.0-21 (2014-01-04) o A new system for user-defined models has been added. o When creating the grid of tuning parameter values, the column names no longer need to be preceded by a period. Periods can still be used as before but are not required. This isn't guaranteed to break backwards compatibility but it may in some cases. o trainControl now has a method = "none" resampling option that bypasses model tuning and fits the model to the entire training set. Note that if more than one model is specified an error will occur. o logicForest models were removed since the package is now archived. o CSimca and RSimca models from the rrcovHD package were added. o Model elm from the elmNN package was added. o Models rknn and rknnBel from the rknn package were added o Model brnn from the brnn package was added. o panel.lift2 and xyplot.lift now have an argument called values that show the percentages of samples found for the specified percentages of samples tested. o train, rfe and sbf should no longer throw a warning that "executing o A ggplot method for train was added. o Imputation via medians was added to preProcess by Zachary Mayer. o A small change was made to rpart models. Previously, when the final model is determined, it would be fit by specifying the model using the cp argument of rpart.control. This could lead to duplicated Cp values in the final list of possible Cp values. The current version fits the final model slightly different. An initial model is fit using cp = 0 then it is pruned using prune.rpart to the desired depth. This shouldn't be different for the vast majority of data sets. Thanks to Jeff Evans for pointing this out. o The method for estimating sigma for SVM and RVM models was slightly changed to make them consistent with how ksvm and rvm does the estimation. o The default behavior for returnResamp in rfeControl and sbfControl is now returnResamp = "final". o cluster was added as a general class with a specific method for resamples objects. o The refactoring of model code resulted in a number of packages being eliminated from the depends field. Additionally, a few were moved to exports. Changes in version 5.17-07 o A bug in spatialSign was fixed for data frames with a single column. o Pre-processing was not applied to the training data set prior to grid creation. This is now done but only for models that use the data when defining the grid. Thanks to Brad Buchsbaum for finding the bug. o Some code was added to rfe to truncate the subset sizes in case the user over-specified them. o A bug was fixed in gamFuncs for the rfe function. o Option in trainControl, rfeControl and sbfControl were added so that the user can set the seed at each resampling iteration (most useful for parallel processing). Thanks to Allan Engelhardt for the recommendation. o Some internal refactoring of the data was done to prepare for some upcoming resampling options. o predict.train now has an explicit na.action argument defaulted to na.omit. If imputation is used in train, then na.action = na.pass is recommended. o A bug was fixed in dummyVars that occured when missing data were in newdata. The function contr.dummy is now deprecated and contr.ltfr should be used (if you are using it at all). Thanks to stackexchange user mchangun for finding the bug. o A check is now done inside dummyVars when levelsOnly = TRUE to see if any predictors share common levels. o A new option fullRank was added to dummyVars. When true, contr.treatment is used. Otherwise, contr.ltfr is used. o A bug in train was fixed with gbm models (thanks to stackoverflow user screechOwl for finding it). Changes in version 5.16-24 (2013-06-20) o The protoclass function in the protoclass package was added. The model uses a distance matrix as input and the train method also uses the proxy package to compute the distance using the Minkowski distance. The two tuning parameters is the neighborhood size (eps) and the Minkowski distance parameter (p). o A bug was (hopefully) fixed that occurred when some type of parallel processing was used with train. The problem is that the methods package was not being loaded in the workers. While reproducible, it is unknown why this occurs and why it is only for some technologies and systems. The methods package is now a formal dependency and we coerce the workers to load it remotely. o A bug was fixed where some calls were printed twice. o For rpart, C5.0 and ksvm, cost-sensitive versions of these models for two classes were added to train. The method values are rpartCost, C5.0Cost and svmRadialWeights. o The prediction code for the ksvm models was changed. There are some cases where the class predictions and the predicted class probabilities disagree. This usually happens when the probabilities are close to 0.50 (in the two class case). A kernlab bug has been filed. In the meantime, if the ksvm model uses a probability model, the class probabilities are generated first and the predicted class is assigned to the probability with the largest value. Thanks to Kjell Johnson for finding that one. o print.train was changed so that tune parameters that are logicals are printed well. Changes in version 5.16-13 o Added a few exemptions to the logic that determines whether a model call should be scrubbed. o An error trap was created to catch issues with missing importance scores in rfe. Changes in version 5.16-03 o A function twoClassSim was added for benchmarking classification models. o A bug was fixed in predict.nullModel related to predicted class probabilities. o The version requirement for gbm was updated. o The function getTrainPerf was made visible. o The automatic tuning grid for sda models from the sda package was changed to include lambda. o When randomForests is used with train and tuneLength == 1, the randomForests default value for mtry is used. o Maximum uncertainty linear discriminant analysis (Mlda) and factor-based linear discriminant analysis (RFlda) from the HiDimDA package were added to train. Changes in version 5.15-87 o Added the Yeo-Johnson power transformation from the car package to the preProcess function. o A train bug was fixed for the rrlda model (found by Tiago Branquinho Oliveira). o The extraTrees model in the extraTrees package was added. o The kknn.train model in the kknn package was added. o A bug was fixed in lrFuncs where the class threshold was improperly set (thanks to David Meyer). o A bug related to newer versions of the gbm package were fixed. Another gbm bug was fixed related to using non-Bernoulli distributions with two class outcomes (thanks to Zachary Mayer). o The old funciton getTrainPerf was finally made visible. o Some models are created using "do.call" and may contain the entire data set in the call object. A function to "scrub" some model call objects was added to reduce their size. o The tuning process for sda:::sda models was changed to add the lambda parameter. Changes in version 5.15-60 o A bug in predictors.earth, discovered by Katrina Bennett, was fixed. o A bug induced by version 5.15-052 for the bootstrap 632 rule was fixed. o The DESCRIPTION file as of 5.15-048 should have used a version-specific lattice dependency. o lift can compute gain and lift charts (and defaults to gain) o The gbm model was updated to handle 3 or more classes. o For bagged trees using ipred, the code in train defaults to keepX = FALSE to save space. Pass in keepX = TRUE to use out-of-bag sampling for this model. o Changes were made to support vector machines for classification models due to bugs with class probabilities in the latest version of kernlab. The prob.model will default to the value of classProbs in the trControl function. If prob.model is passed in as an argument to train, this specification over-rides the default. In other words, to avoid generating a probability model, set either classProbs = FALSE or prob.model = FALSE. Changes in version 5.15-052 (2013-01-17) o Added bayesglm from the arm package. o A few bugs were fixed in bag, thanks to Keith Woolner. Most notably, out-of-bag estimates are now computed when the prediction function includes a column called pred. o Parallel processing was implemented in bag and avNNet, which can be turned off using an optional arguments. o train, rfe, sbf, bag and avNNet were given an additional argument in their respective control files called allowParallel that defaults to TRUE. When Code, the code will be executed in parallel if a parallel backend (e.g. doMC) is registered. When allowParallel = FALSE, the parallel backend is always ignored. The use case is when rfe or sbf calls train. If a parallel backend with P processors is being used, the combination of these functions will create P^2 processes. Since some operations benefit more from parallelization than others, the user has the ability to concentrate computing resources for specific functions. o A new resampling function called createTimeSlices was contributed by Tony Cooper that generates cross-validation indices for time series data. o A few more options were added to trainControl. initialWindow, horizon and fixedWindow are applicable for when method = "timeslice". Another, indexOut is an optional list of resampling indices for the hold-out set. By default, these values are the unique set of data points not in the training set. o A bug was fixed in multiclass glmnet models when generating class probabilities (thanks to Bradley Buchsbaum for finding it). Changes in version 5.15-048 (2012-12-21) o The three vignettes were removed and two things were added: a smaller vignette and a large collection of help pages. o Minkoo Seo found a bug where na.action was not being properly set with train.formula(). o parallel.resamples was changed to properly account for missing values. o Some testing code was removed from probFunction and predictionFunction. o Fixed a bug in sbf exposed by a new version of plyr. o Changed the package dependency on reshape to reshape2. o To be more consistent with recent versions of lattice, the parallel.resamples function was changed to parallelplot.resamples. o Since ksvm now allows probabilities when class weights are used, the default behavior in train is to set prob.model = TRUE unless the user explicitly sets it to FALSE. However, I have reported a bug in ksvm that gives inconsistent results with class weights, so this is not advised at this point in time. o Bugs were fix in predict.bagEarth and predict.bagFDA. o When using rfeControl(saveDetails = TRUE) or sbfControl(saveDetails = TRUE) an additional column is added to object$pred called rowIndex. This indicates the row from the original data that is being held-out. Changes in version 5.15-045 (2012-11-27) o A bug was fixed that induced NA values in SVM model predictions. Changes in version 5.15-042 o Many examples are wrapped in dontrun to speed up cran checking. o The scrda methods were removed from the package (on 6/30/12, R Core sent an email that "since we haven't got fixes for long standing warnings of the rda packages since more than half a year now, we set the package to ORPHANED.") o C50 was added (model codes C5.0, C5.0Tree and C5.0Rules). o Fixed a bug in train with NaiveBayes when fL != 0 was used o The output of train with verboseIter = TRUE was modified to show the resample label as well as logging when the worker started and stopped the task (better when using parallel processing). o Added a long-hidden function downSample for class imbalances o An upSample function was added for class imbalances. o A new file, aaa.R, was added to be compiled first that tries to eliminate the dreaded 'no visible binding for global variable' false positives. Specific namespaces were used with several functions for avoid similar warnings. o A bug was fixed with icr.formula that was so ridiculous, I now know that nobody has ever used that function. o Fixed a bug when using method = "oob" with train o Some exceptions were added to plot.train so that some tuning parameters are better labeled. o dotplot.resamples and bwplot.resamples now order the models using the first metric. o A few of the lattice plots for the resamples class were changed such that when only one metric is shown: the strip is not shown and the x-axis label displays the metric o When using trainControl(savePredictions = TRUE) an additional column is added to object$pred called rowIndex. This indicates the row from the original data that is being held-out. o A variable importance function for nnet objects was created based on Gevrey, M., Dimopoulos, I., & Lek, S. (2003). Review and comparison of methods to study the contribution of variables in artificial neural network models. ecological modelling, 160(3), 249–264. o The predictor function for glmnet was update and a variable importance function was also added. o Raghu Nidagal found a bug in predict.avNNet that was fixed. o sensitivity and specificity were given an na.rm argument. o A first attempt at fault tolerance was added to train. If a model fit fails, the predictions are set to NA and a warning is issued (eg "model fit failed for Fold04: sigma=0.00392, C=0.25"). When verboseIter = TRUE, the warning is also printed to the log. Resampled performance is calculated on only the non-missing estimates. This can also be done during predictions, but must be done on a model by model basis. Fault tolerance was added for kernlab models only at this time. o lift was modified in two ways. First, cuts is no longer an argument. The function always uses cuts based on the number of unique probability estimates. Second, a new argument called label is available to use alternate names for the models (e.g. names that are not valid R variable names). o A bug in print.bag was fixed. o Class probabilities were not being generated for sparseLDA models. o Bugs were fixed in the new varImp methods for PART and RIPPER o Starting using namespaces for ctree and cforest to avoid conflicts between duplicate function names in the party and partykit package o A set of functions for RFE and logistic regression (lrFuncs) was added. o A bug in train with method="glmStepAIC" was fixed so that direction and other stepAIC arguments were honored. o A bug was fixed in preProcess where the number of ICA components was not specified. (thanks to Alexander Lebedev) o Another bug was fixed for oblique random forest methods in train. (thanks to Alexander Lebedev) Changes in version 5.15-023 (2012-03-20) o The list of models that can accept factor inputs directly was expanded to include the RWeka models, ctree, cforest and custom models. o Added model lda2, which tunes by the number of functions used during prediction. o predict.train allows probability predictions for custom models now (thanks to Peng Zhang) o confusionMatrix.train was updated to use the default confusionMatrix code when norm = "none" and only a single hold-out was used. o Added variable importance metrics for PART and RIPPER in the RWeka package. o vignettes were moved from /inst/doc to /vignettes Changes in version 5.14-023 (2012-02-23) o The model details in ?train was changed to be more readable o Added two models from the RRF package. RRF uses a penalty for each predictor based on the scaled variable importance scores from a prior random forest fit. RRFglobal sets a common, global penalty across all predictors. o Added two models from the KRLS package: krlsRadial and krlsPoly. Both have kernel parameters (sigma and degree) and a common regularization parameter lambda. The default for lambda is NA, letting the krls function estimate it internally. lambda can also be specified via tuneGrid. o twoClassSummary was modified to wrap the call to pROC:::roc in a try command. In cases where the hold-out data are only from one class, this produced an error. Now it generates NA values for the AUC when this occurs and a general warning is issued. o The underlying workflows for train were modified so that missing values for performance measures would not throw an error (but will issue a warning). Changes in version 5.13-037 (2012-02-17) o Models mlp, mlpWeightDecay, rbf and rbfDDA were added from RSNNS. o Functions roc, rocPoint and aucRoc finally met their end. The cake was a lie. o This NEWS file was converted over to Rd format. Changes in version 5.13-020 o lift was expanded into lift.formula for calculating the plot points and xyplot.lift to create the plot. o The package vignettes were altered to stop loading external RData files. o A few match.call changes were made to pass new R CMD check tests. o calibration, calibration.formula and xyplot.calibration were created to make probability calibration plots. o Model types xyf and bdk from the kohonen package were added. o update.train was added so that tuning parameters can be manually set if the automated approach to setting their values is insufficient. Changes in version 5.11-006 o When using method = "pls" in train, the plsr function used the default PLS algorithm ("kernelpls"). Now, the full orthogonal scores method is used. This results in the same model, but a more extensive set of values are calculated that enable VIP calculations (without much of a loss in computational efficient). o A check was added to preProcess to ensure valid values of method were used. o A new method, kernelpls, was added. o residuals and summary methods were added to train objects that pass the final model to their respective functions. o Bugs were fixed that prevented hold-out predictions from being returned. Changes in version 5.11-003 o A bug in roc was found when the classes were completely separable. o The ROC calculations for twoClassSummary and filterVarImp were changed to use the pROC package. This, and other changes, have increased efficiency. For filterVarImp on the cell segmentation data lead to a 54-fold decrease in execution time. For the Glass data in the mlbench package, the speedup was 37-fold. Warnings were added for roc, aucRoc and rocPoint regarding their deprecation. o random ferns (package rFerns) were added o Another sparse LDA model (from the penalizedLDA) was also added Changes in version 5.09-006 (2011-12-07) o Shea X made a suggestion and provided code to improve the speed of prediction when sequential parameters are used for gbm models. o Andrew Ziem suggested an error check with metric = "ROC" and classProbs = FALSE. o Andrew Ziem found a bug in how train obtained earth class probabilities Changes in version 5.09-002 o Fixed a bug which occurred when plsda models were used with class probabilities o As of 8/15/11, the glmnet function was updated to return a character vector. Because of this, train required modification and a version requirement was put in the package description file. Changes in version 5.08-011 (2011-11-25) o Andrew Ziem found another small bug with parallel processing and train (functions in the caret namespace cannot be found). o Ben Hoffman found a bug in pickSizeTolerance that was fixed. o Jiaye Yu found (and fixed) a bug in getting predictions back from rfe Changes in version 5.07-024 (2011-11-14) o Using saveDetails = TRUE in sbfControl or rfeControl will save the predictions on the hold-out sets (Jiaye Yu wins the prize for finding that one). o trainControl now has a logical to save the hold-out predictions. Changes in version 5.07-005 (2011-11-08) o type = "prob" was added for avNNet prediction. o A warning was added when a model from RWeka is used with train and (it appears that) multicore is being used for parallel processing. The session will crash, so don't do that. o A bug was fixed where the extrapolation limits were being applied in predict.train but not in extractPrediction. Thanks to Antoine Stevens for finding this. o Modifications were made to some of the workflow code to expose internal functions. When parallel processing was used with doMPI or doSMP, foreach did not find some caret internals (but doMC did). Changes in version 5.07-001 (2011-10-23) o changed calls to predict.mvr since the pls package now has a namespace. Changes in version 5.06-002 o a beta version of custom models with train is included. The "caretTrain" vignette was updated with a new section that defines how to make custom models. Changes in version 5.05-004 o laying some of the groundwork for custom models o updates to get away from deprecated (mean and sd on data frames) o The pre-processing in train bug of the last version was not entirely squashed. Now it is. Changes in version 5.04-007 (2011-09-27) o panel.lift was moved out of the examples in ?lift and into the package along with another function, panel.lift2. o lift now uses panel.lift2 by default o Added robust regularized linear discriminant analysis from the rrlda package o Added evtree from evtree o A weird bug was fixed that occurred when some models were run with sequential parameters that were fixed to single values (thanks to Antoine Stevens for finding this issue). item Another bug was fixed where pre-processing with train could fail Changes in version 5.03-003 (2011-09-24) o pre-processing in train did not occur for the final model fit Changes in version 5.02-011 (2011-09-19) o A function, lift, was added to create lattice objects for lift plots. o Several models were added from the obliqueRF package: 'ORFridge' (linear combinations created using L2 regularization), 'ORFpls' (using partial least squares), 'ORFsvm' (linear support vector machines), and 'ORFlog' (using logistic regression). As of now, the package only support classification. o Added regression models simpls and widekernelpls. These are new models since both train and plsr have an argument called method, so the computational algorithm could not be passed through using the three dots. o Model rpart was added that uses cp as the tuning parameter. To make the model codes more consistent, rpart and ctree correspond to the nominal tuning parameters (cp and mincriterion, respectively) and rpart2 and ctree2 are the alternate versions using maxdepth. o The text for ctree's tuning parameter was changed to '1 - P-Value Threshold' o The argument controls was not being properly passed through in models ctree and ctree2. Changes in version 5.01-001 (2011-09-02) o controls was not being set properly for cforest models in train o The print methods for train, rfe and sbf did not recognize LOOCV o avNNet sometimes failed with categorical outcomes with bag = FALSE o A bug in preProcess was fixed that was triggered by matrices without dimnames (found by Allan Engelhardt) o bagged MARS models with factor outcomes now work o cforest was using the argument control instead of controls o A few bugs for class probabilities were fixed for slda, hdda, glmStepAIC, nodeHarvest, avNNet and sda o When looping over models and resamples, the foreach package is now being used. Now, when using parallel processing, the caret code stays the same and parallelism is invoked using one of the "do" packages (eg. doMC, doMPI, etc). This affects train, rfe and sbf. Their respective man pages have been revised to illustrate this change. o The order of the results produced by defaultSummary were changed so that the ROC AUC is first o A few man and C files were updated to eliminate R CMD check warnings o Now that we are using foreach, the verbose option in trainControl, rfeControl and sbfControl are now defaulted to FALSE o rfe now returns the variable ranks in a single data frame (previously there were data frames in lists of lists) for each of use. This will will break code from previous versions. The built-in RFE functions were also modified o confusionMatrix methods for rfe and sbf were added o NULL values of 'method' in preProcess are no longer allowed o a model for ridge regression was added (method = 'ridge') based on enet. Changes in version 4.98 (2011-07-26) o A bug was fixed in a few of the bagging aggregation functions (found by Harlan Harris). o Fixed a bug spotted by Richard Marchese Robinson in createFolds when the outcome was numeric. The issue is that createFolds is trying to randomize n/4 numeric samples to k folds. With less than 40 samples, it could not always do this and would generate less than k folds in some cases. The change will adjust the number of groups based on n and k. For small samples sizes, it will not use stratification. For larger data sets, it will at most group the data into quartiles. o A function confusionMatrix.train was added to get an average confusion matrices across resampled hold-outs when using the train function for classification. o Added another model, avNNet, that fits several neural networks via the nnet package using different seeds, then averages the predictions of the networks. There is an additional bagging option. o The default value of the 'var' argument of bag was changed. o As requested, most options can be passed from train to preProcess. The trainControl function was re-factored and several options (e.g. k, thresh) were combined into a single list option called preProcOptions. The default is consistent with the original configuration: preProcOptions = list(thresh = 0.95, ICAcomp = 3, k = 5) o nother option was added to preProcess. The pcaComp option can be used to set exactly how many components are used (as opposed to just a threshold). It defaults to NULL so that the threshold method is still used by default, but a non-null value of pcaComp over-rides thresh. o When created within train, the call for preProcess is now modified to be a text string ("scrubed") because the call could be very large. o Removed two deprecated functions: applyProcessing and processData. o A new version of the cell segmentation data was saved and the original version was moved to the package website (see segmentationData for location). First, several discrete versions of some of the predictors (with the suffix "Status") were removed. Second, there are several skewed predictors with minimum values of zero (that would benefit from some transformation, such as the log). A constant value of 1 was added to these fields: AvgIntenCh2, FiberAlign2Ch3, FiberAlign2Ch4, SpotFiberCountCh4 and TotalIntenCh2. Changes in version 4.92 (2011-07-06) o Some tweaks were made to plot.train in a effort to get the group key to look less horrid. o train, rfe and sbf are now able to estimate the time that these models take to predict new samples. Their respective control objects have a new option, timingSamps, that indicates how many of the training set samples should be used for prediction (the default of zero means do not estimate the prediction time). o xyplot.resamples was modified. A new argument, what, has values: "scatter" plots the resampled performance values for two models; "BlandAltman" plots the difference between two models by the average (aka a MA plot) for two models; "tTime", "mTime", "pTime" plot the total model building and tuning; time ("t") or the final model building time ("m") or the time to produce predictions ("p") against a confidence interval for the average performance. 2+ models can be used. o Three new model types were added to train using regsubsets in the leaps package: "leapForward", "leapBackward" and "leapSeq". The tuning parameter, nvmax, is the maximum number of terms in the subset. o The seed was accidentally set when preProcess used ICA (spotted by Allan Engelhardt) o preProcess was always being called (even to do nothing) (found by Guozhu Wen) Changes in version 4.91 (2011-06-09) o Added a few new models associated with the bst package: bstTree, bstLs and bstSm. o A model denoted as "M5" that combines M5P and M5Rules from the RWeka package. This new model uses either of these functions depending on the tuning parameter "rules". Changes in version 4.90 (2011-06-02) o Fixed a bug with train and method = "penalized". Thanks to Fedor for finding it. Changes in version 4.89 (2011-05-28) o A new tuning parameter was added for M5Rules controlling smoothing. o The Laplace correction value for Naive Bayes was also added as a tuning parameter. o varImp.RandomForest was updated to work. It now requires a recent version of the party package. Changes in version 4.88 (2011-04-30) o A variable importance method was created for Cubist models. Changes in version 4.87 (2011-04-28) o Altered the earth/MARS/FDA labels to be more exact. o Added cubist models from the Cubist package. o A new option to trainControl was added to allow users to constrain the possible predicted values of the model to the range seen in the training set or a user-defined range. One-sided ranges are also allowed. Changes in version 4.85 (2011-04-03) o Two typos fixed in print.rfe and print.sbf (thanks to Jan Lammertyn) Changes in version 4.83 (2011-04-01) o dummyVars failed with formulas using "." (all.vars does not handle this well) o tree2 was failing for some classification models o When SVM classification models are used with class.weights, the options prob.model is automatically set to FALSE (otherwise, it is always set to TRUE). A warning is issued that the model will not be able to create class probabilities. o Also for SVM classification models, there are cases when the probability model generates negative class probabilities. In these cases, we assign a probability of zero then coerce the probabilities to sum to one. o Several typos in the help pages were fixed (thanks to Andrew Ziem). o Added a new model, svmRadialCost, that fits the SVM model and estimates the sigma parameter for each resample (to properly capture the uncertainty). o preProcess has a new method called "range" that scales the predictors to [0, 1] (which is approximate for new samples if the training set ranges is narrow in comparison). o A check was added to train to make sure that, when the user passes a data frame to tuneGrid, the names are correct and complete. o print.train prints the number of classes and levels for classification models. Changes in version 4.78 (2011-02-09) o Added a few bagging modules. See ?bag. o Added basic timings of the entire call to train, rfe and sbf as well as the fit time of the final model. These are stored in an element called "times". o The data files were updated to use better compression, which added a higher R version dependency. o plot.train was pretty much re-written to more effectively use trellis theme defaults and to allow arguments (e.g. axis labels, keys, etc) to be passed in to over-ride the defaults. o Bug fix for lda bagging function o Bug fix for print.train when preProc is NULL o predict.BoxCoxTrans would go all klablooey if there were missing values o varImp.rpart was failing with some models (thanks to Maria Delgado) Changes in version 4.77 (2011-01-23) o A new class was added or estimating and applying the Box-Cox transformation to data called BoxCoxTrans. This is also included as an option to transform predictor variables. Although the Box-Tidwell transformation was invented for this purpose, the Box-Cox transformation is more straightforward, less prone to numerical issues and just as effective. This method was also added to preProcess. o Fixed mis-labelled x axis in plot.train when a transformation is applied for models with three tuning parameters. o When plotting a train object with method == "gbm" and multiple values of the shrinkage parameter, the ordering of panels was improved. o Fixed bugs for regression prediction using partDSA and qrf. o Another bug, reported by Jan Lammertyn, related to extractPrediciton with a single predictor was also fixed. Changes in version 4.76 (2011-01-08) o Fixed a bug where linear SVM models were not working for classification Changes in version 4.75 (2011-01-03) o 'gcvEearth' which is the basic MARS model. The pruning procedure is the nominal one based on GCV; only the degree is tuned by train. o 'qrnn' for quantile regression neural networks from the qrnn package. o 'Boruta' for random forests models with feature selection via the Boruta package. Changes in version 4.74 o Some changes to print.train: the call is not automatically printed (but can be when print.train is explicitly invoked); the "Selected" column is also not automatically printed (but can be); non-table text now respects options("width"); only significant digits are now printed when tuning parameters are kept at a constant value Changes in version 4.73 (2010-12-22) o Bug fixes to preProcess related to complete.cases and a single predictor. o For knn models (knn3 and knnreg), added automatic conversion of data frames to matrices Changes in version 4.72 (2010-12-18) o A new function for rfe with gam was added. o "Down-sampling" was implemented with bag so that, for classification models, each class has the same number of classes as the smallest class. o Added a new class, dummyVars, that creates an entire set of binary dummy variables (instead of the reduced, full rank set). The initial code was suggested by Gabor Grothendieck on R-Help. The predict method is used to create dummy variables for any data set. o Added R2 and RMSE functions for evaluating regression models o varImp.gam failed to recognize objects from mgcv o a small fix to test a logical vector filterVarImp o When diff.resamples calculated the number of comparisons, the "models" argument was ignored. o predict.bag was ignoring type = "prob" o Minor updates to conform to R 2.13.0 Changes in version 4.70 (2010-11-10) o Added a warning to train when class levels are not valid R variable names. o Fixed a bug in the variable importance function for multinom objects. o Added p-value adjustments to summary.diff.resamples. Confidence intervals in dotplot.diff.resamples are adjusted accordingly if the Bonferroni is used. o For dotplot.resamples, no point was plotted when the upper and/or lower interval values were NaN. Now, the point is plotted but without the interval bars. o Updated print.rfe to correctly describe new resampling methods. Changes in version 4.69 (2010-10-29) o Fixed a bug in predict.rfe where an error was thrown even though the required predictors were in newdata. o Changed preProcess so that centering and scaling are both automatic when PCA or ICA are requested. Changes in version 4.68 (2010-10-23) o Added two functions, checkResamples and checkConditionalX that identify predictor data with degenerate distributions when conditioned on a factor. o Added a high content screening data set (segmentedData) from Hill et al. Impact of image segmentation on high-content screening data quality for SK-BR-3 cells. BMC bioinformatics (2007) vol. 8 (1) pp. 340. o Fixed bugs in how sbf objects were printed (when using repeated CV) and classification models with earth and classProbs = TRUE. Changes in version 4.67 (2010-10-14) o Added predict.rfe o Added imputation using bagged regression trees to preProcess. o Fixed bug in varImp.rfe that caused incorrect results (thanks to Lawrence Mosley for the find). Changes in version 4.65 (2010-10-09) o Fixed a bug where train would not allow knn imputation. o filterVarImp and roc now check for missing values and use complete data for each predictor (instead of case- wise deletion across all predictors). Changes in version 4.64 (2010-10-07) o Fixed bug introduced in the last version with createDataPartition(... list = FALSE). o Fixed a bug predicting class probabilities when using earth/glm models o Fixed a bug that occurred when train was used with ctree or tree2 methods. o Fixed bugs in rfe and sbf when running in parallel; not all the resampling results were saved Changes in version 4.63 (2010-09-29) o A p-value from McNemar's test was added to confusionMatrix. o Updated print.train so that constant parameters are not shown in the table (but a note is written below the table instead). Also, the output was changed slightly to be more easily read (I hope) o Adapted varImp.gam to work with either mgcv or gam packages. o Expanded the tuning parameters for lvq. o Some of the examples in the Model Building vignette were changed o Added bootstrap 632 rule and repeated cross-validation to trainControl. o A new function, createMultiFolds, is used to generate indices for repeated CV. o The various resampling functions now have *named* lists as output (with prefixes "Fold" for cv and repeated cv and "Resample" otherwise) o Pre-processing has been added to train with the preProcess argument. This has been tested when caret function are used with rfe and sbf (via caretFuncs and caretSBF, respectively). o When preProcess(method = "spatialSign"), centering and scaling is done automatically too. Also, a bug was fixed that stopped the transformation from being executed. o knn imputation was added to preProcess. The RANN package is used to find the neighbors (the knn impute function in the impute library was consistently generating segmentation faults, so we wrote our own). o Changed the behavior of preProcess in situations where scaling is requested but there is no variation in the predictor. Previously, the method would fail. Now a warning is issued and the value of the standard deviation is coerced to be one (so that scaling has no effect). Changes in version 4.62 (2010-09-10) o Added gam from mgcv (with smoothing splines and feature selection) and gam from gam (with basic splines and loess) smoothers. For these models, a formula is derived from the data where "near zero variance" predictors (see nearZerVar) are excluded and predictors with less than 10 distinct values are entered as linear (i.e. unsmoothed) terms. Changes in version 4.61 (2010-09-06) o Changed earth fit for classification models to use the glm argument with a binomial family. o Added varImp.multinom, which is based on the absolute values of the model coefficients Changes in version 4.60 (2010-09-03) o The feature selection vignette was updated slightly (again). Changes in version 4.59 (2010-09-01) o Updated rfe and sbf to include class probabilities in performance calculations. o Also, the names of the resampling indices were harmonized across train, rfe and sbf. o The feature selection vignette was updated slightly. Changes in version 4.58 (2010-08-29) o Added the ability to include class probabilities in performance calculations. See trainControl and twoClassSummary. o Updated and restructured the main vignette. Changes in version 4.57 (2010-08-26) o Internal changes related to how predictions from models are stored and summarized. With the exception of loo, the model performance values are calculated by the workers instead of the main program. This should reduce i/o and lay some groundwork for upcoming changes. o The default grid for relaxo models were changed based on and initial model fit. o partDSA model predictions were modified; there were cases where the user might request X partitions, but the model only produced Y < X. In these cases, the partitions for missing models were replaced with the largest model that was fit. o The function modelLookup was put in the namespace and a man file was added. o The names of the resample indices are automatically reset, even if the user specified them. Changes in version 4.56 o Fixed a bug generated a few versions ago where varImp for plsda and fda objects crashed. Changes in version 4.55 o When computing the scale parameter for RBF kernels, the option to automatically scale the data was changed to TRUE Changes in version 4.54 (2010-08-18) o Added logic.bagging in logicFT with method = "logicBag" Changes in version 4.53 (2010-08-17) o Fixed a bug in varImp.train related to nearest shrunken centroid models. o Added logic regression and logic forests Changes in version 4.51 (2010-08-11) o Added an option to splom.resamples so that the variables in the scatter plots are models or metrics. Changes in version 4.50 o Added dotplot.resamples plus acknowledgements to Hothorn et al. (2005) and Eugster et al. (2008) Changes in version 4.49 (2010-08-09) o Enhanced the tuneGrid option to allow a function to be passed in. Changes in version 4.48 (2010-08-08) o Added a prcomp method for the resamples class Changes in version 4.47 (2010-08-05) o Extended resamples to work with rfe and sbf Changes in version 4.46 o Cleaned up some of the man files for the resamples class and added parallel.resamples. o Fixed a bug in diff.resamples where ... were not being passed to the test statistic function. o Added more log messages in train when running verbose. o Added the German credit data set. Changes in version 4.45 (2010-07-28) o Added a general framework for bagging models via the bag function. Also, model type "hdda" from the HDclassif package was added. Changes in version 4.44 (2010-07-24) o Added neuralnet, quantregForest and rda (from rda) to train. Since there is a naming conflict with rda from mda, the rda model was given a method value of "scrda". Changes in version 4.43 (2010-07-01) o Tthe resampling estimate of the standard deviation given by train since v 4.39 was wrong o A new field was added to varImp.mvr called "estimate". In cases where the mvr model had multiple estimates of performance (e.g. training set, CV, etc) the user can now select which estimate they want to be used in the importance calculation (thanks to Sophie Bréand for finding this) Changes in version 4.42 (2010-06-11) o Added predict.sbf and modified the structure of the sbf helper functions. The "score" function only computes the metric used to filter and the filter function does the actual filtering. This was changed so that FDR corrections or other operations that use all of the p-values can be computed. o Also, the formatting of p-values in print.confusionMatrix was changed o An argument was added to maxDissim so that the variable name is returned instead of the index. o Independent component analysis was added to the list of pre-processing operations and a new model ("icr") was added to fit a pcr-like model with the ICA components. Changes in version 4.40 o Added hda and cleaned up the caret training vignette Changes in version 4.39 (2010-05-16) o Added several classes for examining the resampling results. There are methods for estimating pair-wise differences and lattice functions for visualization. The training vignette has a new section describing the new features. Changes in version 4.38 o Added partDSA and stepAIC for linear models and generalized linear models Changes in version 4.37 (2010-04-19) o Fixed a new bug in how resampling results are exported Changes in version 4.36 (2010-04-18) o Added penalized linear models from the foba package Changes in version 4.35 o Added rocc classification and fixed a typo. Changes in version 4.34 (2010-03-18) o Added two new data sets: dhfr and cars Changes in version 4.33 (2010-03-01) o Added GAMens (ensembles using gams) o Fixed a bug in roc that, for some data cases, would reverse the "positive" class and report sensitivity as specificity and vice-versa. Changes in version 4.32 o Added a parallel random forest method in train using the foreach package. o Also added penalized logistic regression using the plr function in the stepPlr package. Changes in version 4.31 (2009-12-09) o Added a new feature selection function, sbf (for selection by filter). o Fixed bug in rfe that did not affect the results, but did produce a warning. o A new model function, nullModel, was added. This model fits either the mean only model for regression or the majority class model for classification. o Also, ldaFuncs had a bug fixed. o Minor changes to Rd files Changes in version 4.30 (2009-11-09) o For whatever reason, there is now a function in the spls package by the name of splsda that does the same thing. A few functions and a man page were changed to ensure backwards compatibility. Changes in version 4.29 o Added stepwise variable selection for lda and qda using the stepclass function in klaR Changes in version 4.28 o Added robust linear and quadratic discriminant analysis functions from rrcov. o Also added another column to the output of extractProb and extractPrediction that saves the name of the model object so that you can have multiple models of the same type and tell which predictions came from which model. o Changes were made to plotClassProbs: new parameters were added and densityplots can now be produced. Changes in version 4.27 (2009-11-05) o Added nodeHarvest Changes in version 4.26 (2009-10-28) o Fixed a bug in caretFunc that led to NaN variable rankings, so that the first k terms were always selected. Changes in version 4.25 (2009-10-06) o Added parallel processing functionality for rfe Changes in version 4.24 (2009-09-30) o Added the ability to use custom metrics with rfe Changes in version 4.22 o Many Rd changes to work with updated parser. Changes in version 4.21 o Re-saved data in more compressed format Changes in version 4.20 (2009-08-25) o Added pcr as a method Changes in version 4.19 (2009-07-01) o Weights argument was added to train for models that accept weights o Also, a bug was fixed for lasso regression (wrong lambda specification) and other for prediction in naive Bayes models with a single predictor. Changes in version 4.18 (2009-06-18) o Fixed bug in new nearZeroVar and updated format.earth so that it does not automatically print the formula Changes in version 4.17 (2009-06-05) o Added a new version of nearZeroVar from Allan Engelhardt that is much faster Changes in version 4.16 (2009-05-24) o Fixed bugs in extractProb (for glmnet) and filterVarImp. o For glmnet, the user can now pass in their own value of family to train (otherwise train will set it depending on the mode of the outcome). However, glmnet doesn't have much support for families at this time, so you can't change links or try other distributions. Changes in version 4.15 (2009-05-13) o Fixed bug in createFolds when the smallest y value is more than 25 of the data Changes in version 4.14 o Fixed bug in print.train Changes in version 4.13 o Added vbmp from vbmp package Changes in version 4.12 (2009-05-08) o Added additional error check to confusionMatrix o Fixed an absurd typo in print.confusionMatrix Changes in version 4.11 (2009-04-29) o Added: linear kernels for svm, rvm and Gaussian processes; rlm from MASS; a knn regression model, knnreg o A set of functions (class "classDist") to computes the class centroids and covariance matrix for a training set for determining Mahalanobis distances of samples to each class centroid was added o a set of functions (rfe) for doing recursive feature selection (aka backwards selection). A new vignette was added for more details Changes in version 4.10 (2009-03-27) o Added OneR and PART from RWeka Changes in version 4.09 o Fixed error in documentation for confusionMatrix. The old doc had "Detection Prevalence = A/(A+B)" and the new one has "Detection Prevalence =(A+B)(A+B+C+D)". The underlying code was correct. o Added lars (fraction and step as parameters) Changes in version 4.08 (2009-02-18) o Updated train and bagEarth to allow earth for classification models Changes in version 4.07 o Added glmnet models Changes in version 4.06 (2009-01-24) o Added code for sparse PLS classification. o Fix a bug in prediction for caTools::LogitBoost Changes in version 4.05 (2009-01-23) o Updated again for more stringent R CMD check tests in R-devel 2.9 Changes in version 4.04 o Updated for more stringent R CMD check tests in R-devel 2.9 Changes in version 4.03 o Significant internal changes were made to how the models are fit. Now, the function used to compute the models is passed in as a parameter (defaulting to lapply). In this way, users can use their own parallel processing software without new versions of caret. Examples are given in train. o Also, fixed a bug where the MSE (instead of RMSE) was reported for random forest OOB resampling o There are more examples in train. o Changes to confusionMatrix, sensitivity, specificity and the predictive value functions: each was made more generic with default and table methods; confusionMatrix "extractor" functions for matrices and tables were added; the pos/neg predicted value computations were changed to incorporate prevalence; prevalence was added as an option to several functions; detection rate and prevalence statistics were added to confusionMatrix; and the examples were expanded in the help files. o This version of caret will break compatibility with caretLSF and caretNWS. However, these packages will not be needed now and will be deprecated. Changes in version 3.51 (2008-12-05) o Updated the man files and manuals. Changes in version 3.50 o Added qda, mda and pda. Changes in version 3.49 o Fixed bug in resampleHist. Also added a check in the train functions that error trapped with glm models and > 2 classes Changes in version 3.48 o Added glms. Also, added varImp.bagEarth to the namespace. Changes in version 3.47 o Added sda from the sda package. There was a naming conflict between sda::sda and sparseLDA:::sda. The method value for sparseLDA was changed from "sda" to "sparseLDA". Changes in version 3.46 o Added spls from the spls package Changes in version 3.45 (2008-10-18) o Added caching of RWeka objects to that they can be saved to the file system and used in other sessions. (changes per Kurt Hornik on 2008-10-05) Changes in version 3.44 o Added sda from the sparseLDA package (not on CRAN). o Also, a bug was fixed where the ellipses were not passed into a few of the newer models (such as penalized and ppr) Changes in version 3.43 o Added the penalized model from the penalized package. In caret, it is regression only although the package allows for classification via glm models. However, it does not allow the user to pass the classes in (just an indicator matrix). Because of this, it doesn't really work with the rest of the classification tools in the package. Changes in version 3.42 o Added a little more formatting to print.train Changes in version 3.41 o For gbm, let the user over-ride the default value of the distribution argument (brought us by Peter Tait via RHelp). Changes in version 3.40 o Changed predict.preProcess so that it doesn't crash if newdata does not have all of the variables used to originally pre-process *unless* PCA processing was requested. Changes in version 3.39 o Fixed bug in varImp.rpart when the model had only primary splits. o Minor changes to the Affy normalization code o Changed typo in predictors man page Changes in version 3.38 o Added a new class called predictors that returns the names of the predictors that were used in the final model. o Also added ppr from the stats package. o Minor update to the project web page to deal with IE issues Changes in version 3.37 (2008-09-05) o Added the ability of train to use custom made performance functions so that the tuning parameters can be chosen on the basis of things other than RMSE/R-squared and Accuracy/Kappa. o A new argument was added to trainControl called "summaryFunction" that is used to specify the function used to compute performance metrics. The default function preserves the functionality prior to this new version o a new argument to train is "maximize" which is a logical for whether the performance measure specified in the "metric" argument to train should be maximized or minimized. o The selection function specified in trainControl carries the maximize argument with it so that customized performance metrics can be used. o A bug was fixed in confusionMatrix (thanks to Gabor Grothendieck) o Another bug was fixed related to predictions from least square SVMs Changes in version 3.36 o Added superpc from the superpc package. One note: the data argument that is passed to superpc is saved in the object that results from superpc.train. This is used later in the prediction function. Changes in version 3.35 o Added slda from ipred. Changes in version 3.34 o Fixed a few bugs related to the lattice plots from version 3.33. o Also added the ripper (aka JRip) and logistic model trees from RWeka Changes in version 3.33 o Added xyplot.train, densityplot.train, histogram.train and stripplot.train. These are all functions to plot the resampling points. There is some overlap between these functions, plot.train and resampleHist. plot.train gives the average metrics only while these plot all of the resampled performance metrics. resampleHist could plot all of the points, but only for the final optimal set of predictors. o To use these functions, there is a new argument in trainControl called returnResamp which should have values "none", "final" and "all". The default is "final" to be consistent with previous versions, but "all" should be specified to use these new functions to their fullest. Changes in version 3.32 (2008-07-30) o The functions predict.train and predict.list were added to use as alternatives to the extractPrediction and extractProbs functions. o Added C4.5 (aka J48) and rules-based models (M5 prime) from RWeka. o Also added logitBoost from the caTools package. This package doesn't have a namespace and RWeka has a function with the same name. It was suggested to use the "::" prefix to differentiate them (but we'll see how this works).