Changes from FRESA.CAD 3.3.1 to FRESA.CAD version 3.4.2
Enhancements:
Feature Decorrelation changed name now is: GDSTMDecorrelation() and it has been enhanced with fast predictions.
Now it has a predict funtion, and embeded into Filteredfit() funtion.
Univariate filters now work with categorical outcomes
Filteredfit() now removes features with p-values greater than the median p-value.
Rewrite part of the code in latent class modeling HLCM_EM()
Interface Changes:
Now RandomCV uses balanced training as default option
Changes from FRESA.CAD 3.1.1 to FRESA.CAD version 3.3.1
Enhancements:
nearestNeighborImpute now can handle longitudinal datasets
Benchmarking Survival Models: CoxBenchmark
Clustering algorithms: GMVECluster and clusterISODATA
Latent Class Classification: HLCM and HLCM_EM
Hybrid Learning methods: ClustClass and GMVEBSWiMS
Longitudinal Features: trajectoriesPolyFeatures
Train prediction Wrappers for: BeSS and glmnet objects
featureDecorrelation for the estimation of an outcome-steered decorrelated dataframe
filteredFit with normalization and PCA
Bugs:
-Try added to prediction, to avoid crashing errors.
- Calls to epiR::epi.tests updated
Minor Changes:
-FRESAScale OrderLogit function nows retunrs values form -2 to 2.
-ENS changed to ensemble
-Removed par(mfrow = c(1,1)) from code, so user has the freedom to collate plots
Requested changes.
-call to wilcox_free() has been added to cpp code.
Changes from FRESA.CAD 3.0.1 to FRESA.CAD version 3.1.1
Enhancements:
BSWiMS now can do ordinal regression
New functions for cross-validation regression/ordinal and binary classification methods
-randomCV
-RegresionBenchmark
-BinaryBenchmark
-OrdinalBenchmark
-predictStats
Functions for filtering features
-univariate_Logit
-univariate_residual
-univariate_tstudent
-univariate_Wilcoxon
-univariate_correlation
-classic mRMR
Plots:
-barPlotCiError
Interface Changes:
forward selections now requires that the user specify the original number of features for FDR adjustment.
FDR BH p-value correction added into the forward selection algorithms: ForwardSelection.Model.Bin.R and ForwardSelection.Model.Res.R
The adjusted p-value is used at each bootstrap to detect which features can be added to the forward models
Code Changes:
Forward selection now is faster for highly dimensional data sets (n>200 features)
Before finding the optimal feature to add, it ranks features according to the correlation to the current residual.
Other:
Vignettes added
OPENMP removed from Makevars
Changes from FRESA.CAD 3.0.0 to FRESA.CAD version 3.0.1
outcheck.txt was removed from the package.
Changes from FRESA.CAD 2.2.1 to FRESA.CAD version 3.0.0
FRESA.CAD now has the ability to create gene signatures and a simple interface for BSWiMS model generation.
The BSWiMS modeling returns a bagged model after finding a set of candidate models.
Enhancements:
New functions:
-BSWiMS function added
-getSignature function added
-eB:SWIMS functionality
-nearestNeighborImpute for data imputation
Software improvements:
-c++ source code reviewed for efficiency
-bootstrappValidation Variable elimination functions rewritten. Now they attempt to provide models at optimal performance.
-CrossValidation now is using a reduced data set.
Interface Changes:
medianPredict now is called ensemblePredict
Loop_threshold removed from FRESA.Model and cross-validation
fast=FALSE in modelFitting changed to fitFRESA=TRUE
Bugs:
Several bugs corrected.
CRAN:
Removed register from c++ code
Changes from FRESA.CAD 2.2.0 to FRESA.CAD version 2.2.1
Enhancements:
meatMaps(...)
Class color bar next to categories
UnivariateRankVariables()
Now has the option to include only tail analysis
Now it store the beta coefficient
rankInverseNormalDataFrame(...,strata=NA)
Now you can specify a conditional ranking by specifying the strata
Now we can predict results from LASSO by retuning the filtered features
LASSO formulas now reported.
Bugs:
Univariate analysis of improved residuals fixed
several minors bug fixed.
Changes from FRESA.CAD 2.1.3 to FRESA.CAD version 2.2.0
FRESA.CAD expanded its capabilities.
Now it provided Bagged models and ensemble analysis from the list of formulas created by:
+ ForwardSelections.Models.Bin
+ ForwardSelections.Models.Res
+ crossValidationFeatureSelection.Bin
+ crossValidationFeatureSelection.Res
The baggedModel function bag the formula coefficients and creates a single model from the list of formulas.
The plotModels.ROC ensemble the model predictions and creates an unique test evaluation of the ensembled models.
-Added function:
baggedModel.R function for coefficient bagging, and variable frequency analysis
-Enhanced function:
plotModels.ROC.R: This function provides ensemble predictions and confusion analysis table
heatMaps.R can accept a list of five colors for its display
-c++ code revised and minor bugs corrected.
-r code revised and bugs corrected.
Changes from FRESA.CAD 2.0.1 to FRESA.CAD version 2.1.3
FRESA.CAD suffered mayor changes from the previous version.
The new version is more effective in handling memory, some functions and outputs
were renamed.
-Function Name Changes
+backVarElimination to backVarElimination_Bin
+backVarNeRiElimination to backVarElimination_Res
+bootstrapValidation to bootstrapValidation_Bin
+bootstrapNeRiValidation to bootstrapValidation_Res
+bootstrapVarElimination to bootstrapVarEliminiation_Bin
+bootstrapVarNeRiElimination to bootstrapVarEliminiation_Res
+crossValidationFeatureSelection to +crossValidationFeatureSelection_Bin
+crossValidationNeRiFeatureSelection to +crossValidationFeatureSelection_Res
+ReclassificationFRESA.Model to ForwardModel_Res
+NeRIBasedFRESA.Model to ForwardModel_Res
+getVarReclassification to getVar_Bin
+getVarNeRI to getVar_Res
+plot.bootstrapValidation to plot.bootstrapValidation_Bin
+plot.bootstrapValidationNeRI to plot.bootstrapValidation_Res
+updateModel to updateModel_Bin
+updateNeRImodel to updateModel_Res
-Renamed Outputs
Model created form forward models followed by back elimination renamed BSWiMS models
enet renamed LASSO
-Enhancements
+cross-validation now stores ID of sampled subject as well as training fits
+cross-validation now reports the ensemble estimations
+Update model added model-size-based Benjaminiâ€“Hochberg procedure (BH)
+Timeseriesanalysis changed the presentation of p values to t values
+beforeFSC formulas produced before the BH correction
+Minor bugs:
+report equivalent variables for regression models
+removed first term of formula list of cross-validation process
+removed exact wilcoxon test
+other minor bugs
-Code reviews
+NAN were replaced by nan("") c++ function
+cpp code revision to remove abs and sign warnings
+median predict revised to work with the new structure of the formula list provided by cross-validation
+speedglm removed from dependencies
Changes from FRESA.CAD 2.0.1 to FRESA.CAD version 2.0.2
+variable _X from code was renamed _xmat
Changes from FRESA.CAD 2.0 to FRESA.CAD version 2.0.1
C++ code was reviewed to met section 1.6.4 "Portable C and C++ code" of "Writing R Extensions" manual.
Dependencies to c-standard libraries removed and round(x) changed to R::fprec(x,0).
Changes from FRESA.CAD 1.0 to FRESA.CAD version 2.0
+ Added c++ libraries to speed-up feature selection.
The c++ libraries functions were written using ARMADILLO and openMP.
+ FRESAcommons.cpp : Auxiliary functions with ARMADILLO implementations of COX, logit and linear fitting
+ binaryFeatureSelectionCpp.cpp: Main functions for bootstrapping, selection and estimation of
features confidence intervals for binary classification models.
+ regresionFeatureSelectionCpp.cpp: Main functions for bootstrapping, selection and estimation of
linear models coefficients.
+ rankInverseNormalCpp.cpp: Function to standardize features based on their ranking
+ Improvements and bug-fix across the FRESA.CAD package to deal with exceptions and zero size models.
+ Interface changes:
*in bootstrapValidation_Bin(...,dataframe,...)
"dataframe" argument renamed "data"
*in bootstrapValidation_Res(...,dataframe,...)
"dataframe" argument renamed "data"
*in bootVarNeRIElimination(...,bootLoops=64,bootFraction=1.0,...)
"bootLoops" and "bootFraction" arguments renamed "loops" and "fraction" respectively.
*in crossValidationFeatureSelection_Bin(...,dataframe,...,backBootLoops,...,bootEstimations,...)
"dataframe", "backBootLoops" and "bootEstimations" arguments renamed
"data", "elimination.bootstrap.steps" and "bootstrap.steps" respectively
*in crossValidationFeatureSelection_Res(...,dataframe,...,backBootLoops,...)
"dataframe" and "backBootLoops" arguments renamed
"data" and "elimination.bootstrap.steps" respectively
*in featureAdjustment(...,dataframe,...)
"dataframe" argument renamed "data"
*FRESA.Model(...,k,...)
"k" argument renamed "nk"
*getKNNpredictionFromFormula(modelFormula,...,k,...)
"modelFormula" and "k" arguments renamed "model.formula" and "nk" respectively
*getVar.Res(...,dataframe,...)
"dataframe" argument renamed "data"
*getVar.Bin(...,dataframe,...)
"dataframe" argument renamed "data"
*heatMaps(...,dataframe,...)
"dataframe" argument renamed "data"
*listTopCorrelatedVariables(...,dataframe,...)
"dataframe" argument renamed "data"
*ensemblePredict(...,newdata,...,k,...)
"newdata" and "k" arguments renamed "testdata" and "nk" respectively
*modelFitting(model,dataframe,...)
"model" and "dataframe" arguments renamed "mode.formula" and "data" respectively
*ForwardSelection.Model.Res(...,dataframe,...)
"dataframe" argument renamed "data"
*predictForFresa(...,newdata,type,...) ->
"newdata" and "type" arguments renamed "testdata" and "predictType" respectively
*rankInverseNormalDataFrame(varList, dataframe,..)
"varList" and "dataframe" arguments renamed "variableList" and "data" respectively
*ForwardSelection.Model.Bin(...,dataframe,...)
"dataframe" argument renamed "data"
*reportEquivalentVariables(...,dataframe,...)
"dataframe" argument renamed "data"
*residualForFRESA(...,newdata,...)
"newdata" argument renamed "testData"
*timeSerieAnalysis(...,dataframe,...)
"dataframe" argument renamed "data"
*uniRankVar(...,dataframe,...,FitType,..)
"dataframe" and "FitType" arguments renamed "data" and "type" respectively
*univariateRankVariables(...,dataframe,...,FitType,..)
"dataframe" and "FitType" arguments renamed "data" and "type" respectively
*updateModel.Bin(...,dataframe,...)
"dataframe" argument renamed "data"
*updateModel.Res(...,dataframe,...)
"dataframe" argument renamed "data"