NEWS | R Documentation |
Brenton Kenkel added ordered logistic or probit regression to train
using method = "polr"
from MASS
LPH07_1
now encodes the noise variables as binary
Both rfe
and sbf
get arguments for indexOut
for their control functions.
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.
The adaptive mixture discriminant model from the adaptDA package was added as well as a robust mixture discriminant model from the robustDA package.
The multi-class discriminant model using binary predictors in the binda package was added.
Ensembles of partial least squares models (via the enpls) package was added.
A bug using gbm
with Poisson data was fixed (thanks to user eriklampa)
sbfControl
now has a multivariate
option where all the predictors are exposed to the scoring function at once.
A function compare_models
was added that is a simple comparison of models via diff.resamples)
.
The row names for the variables
component of rfe
objects were simplified.
Philipp Bergmeir found a bug that was fixed where bag
would not run in parallel.
predictionBounds
was not implemented during resampling.
A few bug fixes to preProcess
were made related to KNN imputation.
The parameter labels for polynomial SVM models were fixed
The tags for dnn
models were fixed.
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.
A number of changes to comply with section 1.1.3.1 of "Writing R Extensions" were made.
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.
plsRglm
was added.
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.
From the kernlab package, SVM models using string kernels were added: svmBoundrangeString
, svmExpoString
, svmSpectrumString
A function update.rfe
was added.
cluster.resamples
was added to the namespace.
An option to choose the metric
was added to summary.resamples
.
prcomp.resamples
now passed ...
to prcomp
. Also the call to prcomp
uses the formula method so that na.action
can be used.
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.
Several regression simulation functions were added: SLC14_1
, SLC14_2
, LPH07_1
and LPH07_2
print.train
was re-factored so that format.data.frame
is now used. This should behave better when using knitr.
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.
ggplot.train
was modified so that groups are distinguished by color and shape.
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.
A bug was fixed by Jia Xu within the knn imputation code used by preProcess
.
A missing piece of documentation in trainControl
for adaptive models was filled in.
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.
A check was added to trainControl
to make sure that a value of min
makes sense when using adaptive resampling.
A man page with the list of models available via train
was added back into the package. See ?models
.
Thoralf Mildenberger found and fixed a bug in the variable importance calculation for neural network models.
The output of varImp
for pamr
models was updated to clarify the ordering of the importance scores.
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.
A bug was fixed related to how seeds were set inside of train
.
The model "parRF"
(parallel random forest) was added back into the library.
When case weights are specified in train
, the hold-out weights are exposed when computing the summary function.
A check was made to convert a data.table
given to train
to a data frame (see http://stackoverflow.com/questions/23256177/r-caret-renames-column-in-data-table-after-training).
Changes were made that stopped execution of train
if there are no rows in the data (changes suggested by Andrew Ziem)
Andrew Ziem also helped improve the documentation.
Several models were updated to work with case weights.
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.
For some parallel processing technologies, the package now export more internal functions.
A bug was fixed in rfe
that occurred when LOO CV was used.
Another bug was fixed that occurred for some models when
tuneGrid
contained only a single model.
A new system for user-defined models has been added. See http://caret.r-forge.r-project.org/custom_models.html.
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.
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.
logicForest
models were removed since the package is
now archived.
CSimca
and RSimca
models from the rrcovHD
package were added.
Model elm
from the elmNN
package was added.
Models rknn
and rknnBel
from the rknn
package were added
Model brnn
from the brnn
package was added.
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.
train
, rfe
and sbf
should no longer throw
a warning that "executing
A ggplot
method for train
was added.
Imputation via medians was added to preProcess
by Zachary Mayer.
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.
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.
The default behavior for returnResamp
in rfeControl
and
sbfControl
is now returnResamp = "final"
.
cluster
was added as a general class with a specific method
for resamples
objects.
The refactoring of model code resulted in a number of packages being eliminated from the depends field. Additionally, a few were moved to exports.
A bug in spatialSign
was fixed for data frames with
a single column.
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.
Some code was added to rfe
to truncate the subset
sizes in case the user over-specified them.
A bug was fixed in gamFuncs
for the rfe
function.
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.
Some internal refactoring of the data was done to prepare for some upcoming resampling options.
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.
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.
A check is now done inside dummyVars
when
levelsOnly = TRUE
to see if any predictors share common
levels.
A new option fullRank
was added to dummyVars
.
When true, contr.treatment
is used. Otherwise,
contr.ltfr
is used.
A bug in train
was fixed with gbm
models
(thanks to stackoverflow user screechOwl for finding it).
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
).
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.
A bug was fixed where some calls were printed twice.
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
.
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.
print.train
was changed so that tune parameters that are
logicals are printed well.
Added a few exemptions to the logic that determines whether a model call should be scrubbed.
An error trap was created to catch issues with missing importance scores in rfe
.
A function twoClassSim
was added for benchmarking classification models.
A bug was fixed in predict.nullModel
related to predicted class probabilities.
The version requirement for gbm was updated.
The function getTrainPerf
was made visible.
The automatic tuning grid for sda
models from the sda package was changed to include lambda
.
When randomForests
is used with train
and tuneLength == 1
, the randomForests
default value for mtry
is used.
Maximum uncertainty linear discriminant analysis (Mlda
) and factor-based linear discriminant analysis (RFlda
) from the HiDimDA package were added to train
.
Added the Yeo-Johnson power transformation from the car
package to the preProcess
function.
A train
bug was fixed for the rrlda
model (found
by Tiago Branquinho Oliveira).
The extraTrees
model in the extraTrees package was
added.
The kknn.train
model in the kknn package was
added.
A bug was fixed in lrFuncs
where the class threshold was
improperly set (thanks to David Meyer).
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).
The old funciton getTrainPerf
was finally made visible.
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.
The tuning process for sda:::sda
models was changed to
add the lambda
parameter.
A bug in predictors.earth
, discovered by Katrina Bennett,
was fixed.
A bug induced by version 5.15-052 for the bootstrap 632 rule was fixed.
The DESCRIPTION file as of 5.15-048 should have used a version-specific lattice dependency.
lift
can compute gain and lift charts (and defaults to
gain)
The gbm model was updated to handle 3 or more classes.
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.
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
.
Added bayesglm
from the arm package.
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
.
Parallel processing was implemented in bag
and
avNNet
, which can be turned off using an optional arguments.
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.
A new resampling function called createTimeSlices
was
contributed by Tony Cooper that generates cross-validation indices for
time series data.
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.
A bug was fixed in multiclass glmnet
models when
generating class probabilities (thanks to Bradley Buchsbaum for
finding it).
The three vignettes were removed and two things were added: a smaller vignette and a large collection of help pages at http://caret.r-forge.r-project.org/.
Minkoo Seo found a bug where na.action
was not being properly
set with train.formula().
parallel.resamples
was changed to properly account for
missing values.
Some testing code was removed from probFunction
and
predictionFunction
.
Fixed a bug in sbf
exposed by a new version of plyr.
To be more consistent with recent versions of lattice,
the parallel.resamples
function was changed to
parallelplot.resamples
.
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.
Bugs were fix in predict.bagEarth
and
predict.bagFDA
.
When using rfeControl(saveDetails = TRUE)
or
sbfControl(saveDetails = TRUE)
an additional column is
added to object$pred
called rowIndex
. This indicates the
row from the original data that is being held-out.
A bug was fixed that induced NA
values in SVM model predictions.
Many examples are wrapped in dontrun to speed up cran checking.
The scrda
methods were removed from the package (on
6/30/12, R Core sent an email that "since we haven't got fixes for
long standing warnings of the rda packages since more than half a year
now, we set the package to ORPHANED.")
C5.0 was added (model codes C5.0
, C5.0Tree
and
C5.0Rules
).
Fixed a bug in train
with NaiveBayes when fL != 0
was used
The output of train
with verboseIter = TRUE
was
modified to show the resample label as well as logging when the worker
started and stopped the task (better when using parallel processing).
Added a long-hidden function downSample
for class imbalances
An upSample
function was added for class imbalances.
A new file, aaa.R, was added to be compiled first that tries to eliminate the dreaded 'no visible binding for global variable' false positives. Specific namespaces were used with several functions for avoid similar warnings.
A bug was fixed with icr.formula
that was so ridiculous,
I now know that nobody has ever used that function.
Fixed a bug when using method = "oob"
with train
Some exceptions were added to plot.train
so that some
tuning parameters are better labeled.
dotplot.resamples
and bwplot.resamples
now order
the models using the first metric.
A few of the lattice plots for the resamples
class were
changed such that when only one metric is shown: the strip is not
shown and the x-axis label displays the metric
When using trainControl(savePredictions = TRUE)
an
additional column is added to object$pred
called
rowIndex
. This indicates the row from the original data that is
being held-out.
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.
The predictor
function for glmnet
was update and a
variable importance function was also added.
Raghu Nidagal found a bug in predict.avNNet
that was
fixed.
sensitivity
and specificity
were given an
na.rm
argument.
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.
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).
A bug in print.bag
was fixed.
Class probabilities were not being generated for sparseLDA models.
Bugs were fixed in the new varImp methods for PART and RIPPER
Starting using namespaces for ctree
and cforest
to
avoid conflicts between duplicate function names in the party
and partykit package
A set of functions for RFE and logistic regression
(lrFuncs
) was added.
A bug in train
with method="glmStepAIC"
was fixed
so that direction
and other stepAIC
arguments were
honored.
A bug was fixed in preProcess
where the number of ICA
components was not specified. (thanks to Alexander Lebedev)
Another bug was fixed for oblique random forest methods in
train
. (thanks to Alexander Lebedev)
The list of models that can accept factor inputs directly was
expanded to include the RWeka models, ctree
,
cforest
and custom models.
Added model lda2
, which tunes by the number of functions
used during prediction.
predict.train
allows probability predictions for custom
models now (thanks to Peng Zhang)
confusionMatrix.train
was updated to use the default
confusionMatrix
code when norm = "none"
and only a
single hold-out was used.
Added variable importance metrics for PART and RIPPER in the RWeka package.
vignettes were moved from /inst/doc to /vignettes
The model details in ?train
was changed to be more
readable
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.
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
.
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.
The underlying workflows for train
were modified so that
missing values for performance measures would not throw an error (but
will issue a warning).
Models mlp
, mlpWeightDecay
, rbf
and
rbfDDA
were added from RSNNS.
Functions roc
, rocPoint
and aucRoc
finally
met their end. The cake was a lie.
This NEWS file was converted over to Rd format.
lift
was expanded into lift.formula
for calculating the plot points and xyplot.lift
to
create the plot.
The package vignettes were altered to stop loading external RData files.
A few match.call
changes were made to pass new R CMD
check tests.
calibration
, calibration.formula
and
xyplot.calibration
were created to make probability
calibration plots.
Model types xyf
and bdk
from the kohonen
package were added.
update.train
was added so that tuning parameters
can be manually set if the automated approach to setting their
values is insufficient.
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).
A check was added to preProcess
to ensure valid
values of method
were used.
A new method, kernelpls
, was added.
residuals
and summary
methods were added to
train
objects that pass the final model to their
respective functions.
Bugs were fixed that prevented hold-out predictions from being returned.
A bug in roc
was found when the classes were completely
separable.
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.
random ferns (package rFerns) were added
Another sparse LDA model (from the penalizedLDA) was also added
Fixed a bug which occurred when plsda
models were used with class
probabilities
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.
Shea X made a suggestion and provided code to improve the speed
of prediction when sequential parameters are used for
gbm
models.
Andrew Ziem suggested an error check with metric = "ROC"
and
classProbs = FALSE
.
Andrew Ziem found a bug in how train
obtained
earth
class probabilities
Andrew Ziem found another small bug with parallel processing and
train
(functions in the caret namespace cannot be found).
Ben Hoffman found a bug in pickSizeTolerance
that was fixed.
Jiaye Yu found (and fixed) a bug in getting predictions back from
rfe
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).
trainControl
now has a logical to save the hold-out predictions.
type = "prob"
was added for avNNet
prediction.
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.
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.
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).
changed calls to predict.mvr
since the pls package now has a
namespace.
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.
laying some of the groundwork for custom models
updates to get away from deprecated (mean and sd on data frames)
The pre-processing in train
bug of the last
version was not entirely squashed. Now it is.
panel.lift
was moved out of the examples in ?lift
and into the
package along with another function, panel.lift2
.
lift
now uses panel.lift2
by default
Added robust regularized linear discriminant analysis from the rrlda package
Added evtree
from evtree
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
pre-processing in train
did not occur for the final model fit
A function, lift
, was added to create lattice
objects for lift plots.
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.
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.
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
.
The text for ctree
's tuning parameter was changed to '1 -
P-Value Threshold'
The argument controls
was not being properly passed
through in models ctree
and ctree2
.
controls
was not being set properly for cforest
models in train
The print methods for train
, rfe
and
sbf
did not recognize LOOCV
avNNet
sometimes failed with categorical outcomes with bag = FALSE
A bug in preProcess
was fixed that was triggered by matrices without
dimnames (found by Allan Engelhardt)
bagged MARS models with factor outcomes now work
cforest
was using the argument control
instead of controls
A few bugs for class probabilities were fixed for slda
, hdda
,
glmStepAIC
, nodeHarvest
, avNNet
and sda
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.
The order of the results produced by defaultSummary
were changed
so that the ROC AUC is first
A few man and C files were updated to eliminate R CMD check warnings
Now that we are using foreach, the verbose option in trainControl
,
rfeControl
and sbfControl
are now defaulted to FALSE
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
confusionMatrix methods for rfe
and sbf
were added
NULL values of 'method' in preProcess
are no longer allowed
a model for ridge regression was added (method = 'ridge'
) based on enet
.
A bug was fixed in a few of the bagging aggregation functions (found by Harlan Harris).
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.
A function confusionMatrix.train
was added to get an average
confusion matrices across resampled hold-outs when using the
train
function for classification.
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.
The default value of the 'var' argument of bag
was changed.
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)
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
.
When created within train
, the call for preProcess
is now
modified to be a text string ("scrubed") because the call could
be very large.
Removed two deprecated functions: applyProcessing
and
processData
.
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
.
Some tweaks were made to plot.train
in a effort to get the group
key to look less horrid.
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).
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.
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.
The seed was accidentally set when preProcess
used ICA (spotted
by Allan Engelhardt)
preProcess
was always being called (even to do nothing)
(found by Guozhu Wen)
Added a few new models associated with the bst package: bstTree, bstLs and bstSm.
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"
.
Fixed a bug with train
and method = "penalized"
. Thanks to
Fedor for finding it.
A new tuning parameter was added for M5Rules
controlling smoothing.
The Laplace correction value for Naive Bayes was also added as a tuning parameter.
varImp.RandomForest
was updated to work. It now requires a recent
version of the party package.
A variable importance method was created for Cubist models.
Altered the earth/MARS/FDA labels to be more exact.
Added cubist models from the Cubist package.
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.
Two typos fixed in print.rfe
and
print.sbf
(thanks to Jan Lammertyn)
dummyVars
failed with formulas using "."
(all.vars
does not handle this well)
tree2
was failing for some classification models
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.
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.
Several typos in the help pages were fixed (thanks to Andrew Ziem).
Added a new model, svmRadialCost
, that fits the SVM model
and estimates the sigma
parameter for each resample (to
properly capture the uncertainty).
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).
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.
print.train
prints the number of classes and levels for classification
models.
Added a few bagging modules. See ?bag.
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".
The data files were updated to use better compression, which added a higher R version dependency.
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.
Bug fix for lda bagging function
Bug fix for print.train
when preProc
is NULL
predict.BoxCoxTrans
would go all klablooey if there were missing
values
varImp.rpart
was failing with some models (thanks to Maria Delgado)
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
.
Fixed mis-labelled x axis in plot.train
when a
transformation is applied for models with three tuning parameters.
When plotting a train
object with method ==
"gbm"
and multiple values of the shrinkage parameter, the ordering of
panels was improved.
Fixed bugs for regression prediction using partDSA
and
qrf
.
Another bug, reported by Jan Lammertyn, related to
extractPrediciton
with a single predictor was also
fixed.
Fixed a bug where linear SVM models were not working for classification
'gcvEearth'
which is the basic MARS model. The pruning procedure
is the nominal one based on GCV; only the degree is tuned by train
.
'qrnn'
for quantile regression neural networks from the qrnn package.
'Boruta'
for random forests models with feature selection via the
Boruta package.
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
Bug fixes to preProcess
related to complete.cases and a single predictor.
For knn models (knn3 and knnreg), added automatic conversion of data frames to matrices
A new function for rfe
with gam was added.
"Down-sampling" was implemented with bag
so that, for
classification models, each class has the same number of classes
as the smallest class.
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.
Added R2
and RMSE
functions for evaluating regression models
varImp.gam
failed to recognize objects from mgcv
a small fix to test a logical vector filterVarImp
When diff.resamples
calculated the number of comparisons,
the "models"
argument was ignored.
predict.bag
was ignoring type = "prob"
Minor updates to conform to R 2.13.0
Added a warning to train
when class levels are not
valid R variable names.
Fixed a bug in the variable importance function for
multinom
objects.
Added p-value adjustments to
summary.diff.resamples
. Confidence intervals in
dotplot.diff.resamples
are adjusted accordingly if the
Bonferroni is used.
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.
Updated print.rfe
to correctly describe new
resampling methods.
Fixed a bug in predict.rfe
where an error was
thrown even though the required predictors were in newdata
.
Changed preProcess
so that centering and scaling are both automatic
when PCA or ICA are requested.
Added two functions, checkResamples
and
checkConditionalX
that identify predictor data with
degenerate distributions when conditioned on a factor.
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.
Fixed bugs in how sbf
objects were printed (when using repeated
CV) and classification models with earth and classProbs = TRUE
.
Added predict.rfe
Added imputation using bagged regression trees to
preProcess
.
Fixed bug in varImp.rfe
that caused incorrect
results (thanks to Lawrence Mosley for the find).
Fixed a bug where train
would not allow knn imputation.
filterVarImp
and roc
now check for missing values and
use complete data for each predictor (instead of case-
wise deletion across all predictors).
Fixed bug introduced in the last version with
createDataPartition(... list = FALSE)
.
Fixed a bug predicting class probabilities when using earth/glm models
Fixed a bug that occurred when train
was used with
ctree
or tree2
methods.
Fixed bugs in rfe
and sbf
when running in
parallel; not all the resampling results were saved
A p-value from McNemar's test was added to confusionMatrix
.
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)
Adapted varImp.gam
to work with either mgcv or gam packages.
Expanded the tuning parameters for lvq
.
Some of the examples in the Model Building vignette were changed
Added bootstrap 632 rule and repeated cross-validation
to trainControl
.
A new function, createMultiFolds
, is
used to generate indices for repeated CV.
The various resampling functions now have *named* lists as output (with prefixes "Fold" for cv and repeated cv and "Resample" otherwise)
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).
When preProcess(method = "spatialSign")
, centering and
scaling is done automatically too. Also, a bug was fixed
that stopped the transformation from being executed.
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).
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).
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.
Changed earth fit for classification models to use the
glm
argument with a binomial family.
Added varImp.multinom
, which is based on the absolute
values of the model coefficients
The feature selection vignette was updated slightly (again).
Updated rfe
and sbf
to include class probabilities
in performance calculations.
Also, the names of the resampling indices were harmonized
across train
, rfe
and sbf
.
The feature selection vignette was updated slightly.
Added the ability to include class probabilities in
performance calculations. See trainControl
and
twoClassSummary
.
Updated and restructured the main vignette.
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.
The default grid for relaxo models were changed based on and initial model fit.
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.
The function modelLookup
was put in the namespace and
a man file was added.
The names of the resample indices are automatically reset, even if the user specified them.
Fixed a bug generated a few versions ago where varImp
for plsda
and fda
objects crashed.
When computing the scale parameter for RBF kernels, the
option to automatically scale the data was changed to TRUE
Added logic.bagging
in logicFT with method = "logicBag"
Fixed a bug in varImp.train
related to nearest shrunken
centroid models.
Added logic regression and logic forests
Added an option to splom.resamples
so that the variables in the
scatter plots are models or metrics.
Added dotplot.resamples
plus acknowledgements to Hothorn et al
(2005) and Eugster et al (2008)
Enhanced the tuneGrid
option to allow a function
to be passed in.
Added a prcomp
method for the resamples
class
Extended resamples
to work with rfe
and sbf
Cleaned up some of the man files for the resamples class
and added parallel.resamples
.
Fixed a bug in diff.resamples
where ...
were
not being passed to the test statistic function.
Added more log messages in train
when running verbose.
Added the German credit data set.
Added a general framework for bagging models via the
bag
function. Also, model type "hdda"
from the
HDclassif package was added.
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"
.
Tthe resampling estimate of the standard deviation given
by train
since v 4.39 was wrong
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)
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.
Also, the formatting of p-values in print.confusionMatrix
was changed
An argument was added to maxDissim
so that the variable name is returned instead of the index.
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.
Added hda
and cleaned up the caret training vignette
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.
Added partDSA and stepAIC
for linear models and
generalized linear models
Fixed a new bug in how resampling results are exported
Added penalized linear models from the foba package
Added rocc
classification and fixed a typo.
Added two new data sets: dhfr
and cars
Added GAMens (ensembles using gams)
Fixed a bug in roc
that, for some data cases, would reverse the "positive"
class and report sensitivity as specificity and vice-versa.
Added a parallel random forest method in train
using the foreach package.
Also added penalized logistic regression using the plr
function in the
stepPlr package.
Added a new feature selection function, sbf
(for selection by filter).
Fixed bug in rfe
that did not affect the results, but did produce
a warning.
A new model function, nullModel
, was added. This model fits either the
mean only model for regression or the majority class model for classification.
Also, ldaFuncs had a bug fixed.
Minor changes to Rd files
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.
Added stepwise variable selection for lda
and qda
using the
stepclass
function in klaR
Added robust linear and quadratic discriminant analysis functions from rrcov.
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.
Changes were made to plotClassProbs
: new parameters were added
and densityplots can now be produced.
Added nodeHarvest
Fixed a bug in caretFunc
that led to NaN variable rankings, so
that the first k terms were always selected.
Added parallel processing functionality for rfe
Added the ability to use custom metrics with rfe
Many Rd changes to work with updated parser.
Re-saved data in more compressed format
Added pcr
as a method
Weights argument was added to train
for models that accept weights
Also, a bug was fixed for lasso regression (wrong lambda specification) and other for prediction in naive Bayes models with a single predictor.
Fixed bug in new nearZeroVar
and updated format.earth
so that it
does not automatically print the formula
Added a new version of nearZeroVar
from Allan Engelhardt that is
much faster
Fixed bugs in extractProb
(for glmnet) and filterVarImp
.
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.
Fixed bug in createFolds
when the smallest y value is more than 25
of the data
Fixed bug in print.train
Added vbmp from vbmp package
Added additional error check to confusionMatrix
Fixed an absurd typo in print.confusionMatrix
Added: linear kernels for svm, rvm and Gaussian processes; rlm
from MASS; a knn regression model, knnreg
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
a set of functions (rfe
) for doing recursive feature selection
(aka backwards selection). A new vignette was added for more
details
Added OneR
and PART
from RWeka
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.
Added lars
(fraction
and step
as parameters)
Updated train
and bagEarth
to allow earth
for classification models
Added glmnet models
Added code for sparse PLS classification.
Fix a bug in prediction for caTools::LogitBoost
Updated again for more stringent R CMD check tests in R-devel 2.9
Updated for more stringent R CMD check tests in R-devel 2.9
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
.
Also, fixed a bug where the MSE (instead of RMSE) was reported for random forest OOB resampling
There are more examples in train
.
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.
This version of caret will break compatibility with caretLSF and caretNWS. However, these packages will not be needed now and will be deprecated.
Updated the man files and manuals.
Added qda
, mda
and pda
.
Fixed bug in resampleHist
. Also added a check in the train
functions
that error trapped with glm
models and > 2 classes
Added glm
s. Also, added varImp.bagEarth
to the
namespace.
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".
Added spls
from the spls package
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)
Added sda
from the sparseLDA package (not on
CRAN).
Also, a bug was fixed where the ellipses were not passed into a
few of the newer models (such as penalized
and ppr
)
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.
Added a little more formatting to print.train
For gbm
, let the user over-ride the default value of the
distribution
argument (brought us by Peter Tait via RHelp).
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.
Fixed bug in varImp.rpart
when the model had only primary
splits.
Minor changes to the Affy normalization code
Changed typo in predictors
man page
Added a new class called predictors
that returns the
names of the predictors that were used in the final model.
Also added ppr
from the stats package.
Minor update to the project web page to deal with IE issues
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.
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
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.
The selection function specified in trainControl
carries
the maximize argument with it so that customized performance
metrics can be used.
A bug was fixed in confusionMatrix
(thanks to Gabor
Grothendieck)
Another bug was fixed related to predictions from least square SVMs
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.
Added slda
from ipred.
Fixed a few bugs related to the lattice plots from version 3.33.
Also added the ripper (aka JRip
) and logistic model trees
from RWeka
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.
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.
The functions predict.train
and predict.list
were
added to use as alternatives to the extractPrediction
and
extractProbs
functions.
Added C4.5 (aka J48
) and rules-based models (M5 prime) from
RWeka.
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).