Library Layout and Organization

Table of Contents GURLS GURLS Available Tasks GURLS++ GURLS++ Available Tasks bGURLS bGURLS Available Tasks bGURLS++ bGURLS++ Available Tasks

GURLS

The GURLS module is composed of the following files and directories:

gurls.m, implementing the GURLS core
defopt.m, a function used to build a default options' structure
demo, a directory containg the demos and the data
quickandirty, a folder containing 2 functions for fast GURLS usage
utils, a directory containing utility functions
tasks directory and files, where the available tasks are listed in the table below

List of Gurls tasks organized by category
Task category	Description	Available tasks
split	Splits data into one or more pair of training and validation sets	ho
paramsel	Performs selection of the regularization parameter lambda and, if using Gaussian kernel, also of the kernel parameter sigma	fixlambda, loocvprimal, loocvdual, hoprimal, hodual, siglam, siglamho, bfprimal, bfdual, calibratesgd, hoprimalr, hodualr, horandfeats, gpregrLambdaGrid, gpregrSigLambGrid, loogpregr, hogpregr, siglamhogpregr, siglamloogpregr,
kernel	Builds the symmetric kernel matrix to be used for training	chisquared, linear, load, randfeats, rbf
rls	Computes the optimizer	primal, dual, auto, pegasos, primalr, dualr, randfeats, gpregr
predkernel	Builds the train-test kernel matrix	traintest
pred	Predicts the labels	primal, dual, randfeats, gpregr
perf	Assess prediction performance	macroavg, precrec, rmse
conf	Computes a confidence for the highest scoring class	maxscore, gap, boltzmangap, boltzman

The GURLS++ module is composed of the following directories:

include, directory containing the headers of the GURLS++ core and tasks, and the implementation of template classes and methods
src, directory containing the implementation of the non template classes and methods
demo, containing the demos
data, containing examples of data
doc, containing the documentation (after running doxygen)

List of Gurls++ tasks organized by category
Task category	Description	Class name	Available subclasses (tasks)
split	Splits data into one or more pair of training and validation sets	Split	Ho
paramsel	Performs selection of the regularization parameter lambda and, if using Gaussian kernel, also of the kernel parameter sigma	ParamSelection	LoocvDual, LoocvPrimal, HoDual, HoPrimal, SiglamHo, Siglam, FixLambda FixSigLam, HoPrimalr, HoDualr, LooGPRegr, HoGPRegr, SigLamLooGPRegr, SigLamHoGPRegr, CalibrateSGD
kernel	Builds the symmetric kernel matrix to be used for training	Kernel	ChisquaredKernel, LinearKernel, RBFKernel
rls	Computes the optimizer	Optimizer	RLSPrimal, RLSDual, RLSAuto, RLSPegasos, RLSPrimalr, RLSDualr, RLSGPRegr, RLSPrimalRecInit, RLSPrimalRecUpdate
predkernel	Builds the train-test kernel matrix	PredKernel	TrainTest
pred	Predicts the labels	Prediction	PredPrimal, PredDual, PredGPRegr
perf	Assess prediction performance	Performance	MacroAvg, PrecisionRecall, Rmse
conf	Confidence	Computes a confidence for the highest scoring class	ConfMaxScore, ConfGap, ConfBoltzmanGap, ConfBoltzman

The GURLS module is composed of the following files and directories:

bgurls.m, implementing the bGURLS core
bigdefopt.m, a function used to build a default options' structure
demo, a directory containg the demos and the data
utils, a directory containing utility functions
adm, a directory containing utility functions for bigarrays and distributed computing management
tasks directory and files, where the available tasks are listed in the table below

List of bGurls tasks organized by category
Task category	Description	Available tasks
bigparamsel	performs selection of the regularization parameter lambda and	hoprimal, calibratesgd, dhoprimal,
bigrls	Computes the optimizer	primal, pegasos, dprimal
bigpred	Predicts the labels	primal
bigperf	Assess prediction performance	macroavg

The bGURLS++ module is composed of the following directories:

include, directory containing the headers of the bGURLS++ core and tasks, and the implementation of template classes and methods
src, directory containing the implementation of the non template classes and methods
demo, containing the demos
data, containing examples of data
doc, containing the documentation (after running doxygen)

List of bGurls++ tasks organized by category
Task category	Description	Class name	Available subclasses (tasks)
bigsplit	Splits data into one or more pair of training and validation sets	BigSplit	Ho
bigparamsel	performs selection of the regularization parameter lambda	BigParamSelection	HoPrimal
bigrls	Computes the optimizer	BigOptimizer	RLSPrimal
bigpred	Predicts the labels	BigPrediction	PredPrimal
bigperf	BigPerformance	Assess prediction performance	MacroAvg