2018-01-07 Greg Vaughan
* Added stochastic stagewise funcitonality
1) currently only implemented for see function
2) using the sgee.control auxiliary function, the following
parameters can be provided by the user to specify the
stochastic version of see where the full data set is subsampled
to improve scalability: stochastic, sampleProb, reSample, and
withReplacement.
3) the default value of stochastic == 1 implements no subsampling,
i.e. the normal stagewise approach
4) for 0
* Added functionality for binary data
1) Originally, genData function was not set up to produce
"correlated" binary data, now it can
2) similarly, the sgee.summary function treated binary
settings in the same way it treated gaussian data. Now,
the function can evaluate based on classification rate
* Minor corrections
1) corrected a typo in bisee maunal
2) corrected some minor presentations in genData and
print.sgeeSummary functions
2017-07-02 Greg Vaughan
* Corrected multiple issues in isee
1) originally, isee could not create an interractionID
in the presence of more than two factor levels
2) also, isee could not construct interactionIDs properly when there was only one covaraite
3) isee did not check whether or not factor levels were actually obsevred, now it does
4) neither did bisee or hisee, now they do
2017-06-21 Greg Vuaghan
* Merged isee and hlsee into one function called isee
1) also isee and hlsee methods were renamed ACTS and HiLa, repsectively
2017-06-17 Greg Vuaghan
* Corrected estimation of correlation parameter w/ Ar-1 structure
2017-06-06 Greg Vaughan
* Corrected implementation of genData to allows for xVariance
2017-05-16 Greg Vaughan
* Added standardization functionality to isee and hlsee
1) standardization occurs immediately before fitting takes place
2) this is after the intercept terms have been generated, so both the main covariates and the interaction terms are standardized
2017-02-28 Greg Vaughan
* Added intercept estimation switch, interceptLimit
1) default implementation uses uniroot with extendInt to estimate
intercept in each iteration
2) when interceptLimit is set however, a fixed bound on the intercept
can be set
3) This allows compatability with older versions of R which did not
have the extendInt functionality
2017-02-20 Greg Vaughan
* Updated isee
1) changed the implementation of the active set management
2) speed should be improved mow
2017-02-14 Greg Vaughan
* Updated hlsee
1) changed the way the proper update is calculated
2) method of selection is the dsame, calculations are just
done in a faster way
2017-02-12 Greg Vaughan
* Updated hlsee
1) changed decision process when choosing main v. int effects
2) now uses |U_{ij}| + |U_{ii}| + |U_{jj}| vs. 3|U_{zz}|
2017-01-04 Greg Vaughan
* Patched bug affecting the calculation of stopping threshold
1) nrow of y, the response vector was used
2) replaced with length, since y may be a vector
2016-12-27 Greg Vaughan
* Added stepsize reduction functionality
1) using new control variable called undoThreshold
2) if sum(abs(oldDeta + newDelta)) <= undoThreshold, then
stepsize is reduced
3) added functionality to also reduce threshold if >= stepsize
4) Corrected a bug in plot function that would use the range
of the intercept even if the intercept wasn't being plotted
5) add row of zeros in plot function so path starts from origin,
not including the itnercept which simply starts from initial estimate
2016-10-19 Greg Vaughan
* Began adding interaction stagewisee implementations
1) added, isee, which implements zhu's work
2) added hlsee, which implements bien's hierarchical lasso
3) modified genData to allow for the creation of interaction data
2016-10-16 Greg Vaughan
* Updated various functions
1) added subset and contrasts functionality
2) added a print function for sgee objects
3) other minor changes/ polishes
2016-09-15 Greg Vaughan
* Began preparing package for CRAN submission
1) Re-did comments the were previously done manually using roxygen2
2) Removed older functiosn that were not intended for distirbution
3) Renamed packaged to sgee
2016-07-26 Greg Vaughan
* Added Poisson data generation
1) simpleDataGeneration function can now generate correlated
poisson data
2016-02-20 Greg Vaughan
* Added AUCCalculator function
1) calculates AUC for a set of logisitc regression fittings
vs. true values
* Cleaned SEE, GSEE, and examples in manual files
* Removed Dependancy on threeboost and MAtrix
2016-02-19 Greg Vaughan
* Added analyzeCoefficientPath function
1) takes in a coefficient path, and a predictive dataset
and analyzes it.
2) if given the true coefficent values, yields additional measures
* Added coefTraceplot function
1) plots the coefficient traceplot
2) provides options for yielding color or black and white plots
3) allows for covariate grouping identification
2016-02-17 Greg Vaughan
* Added simpleDataGeneration, a function to generate gaussian
Data to test stagewise techniques
1) generates gaussian Data that is correlated
2) generates covariates that are also correlated
* Removed older implemnetations of SEE, BiSEE, and GSEE
2016-01-13 Greg Vaughan
* Added GSEE, a simpler implementation of gBoost
1) Execute the forward stagewise procedure
2) uses L2 norm (the group lasso) as penalty
3)generates path as in the other functions
2015-12-29 Greg Vaughan
* Added SEE, a simple implementation of EEBoost using
1) Execute the forward stagewise procedure
2) uses absolute value function (the lasso) as penalty
3)generates path as in the other functions
2015-12-05 Greg Vaughan
* Modified How hBoost and sgBoost are executed
1) evaluateGEE.R was changed so that the intercept was updated
2) then phi and alpha are updated,
3) then the working correlation matrix is updated using updated alpha
4) then the mean parameters estimating equations are calculated
* Cleaned some of the manual files
2015-09-20 Greg Vaughan
* Modified Package structure
1) primary functions are called "hBoost" and
"sgBoost", for the hBoost and sgboost techniques.
2) other functions still present for simulation
purposes, but most will eventually be removed
* Added Examples for hBoost and sgBoost
2015-08-17 Greg Vaughan
* Added gBoost, sgBoost, and hBoost
1) gBoost is threeBoostGroup but estimates
nuisance parameters as per liang and Zeger
(and under the assumption of one of each)
2) sgBoost is threeBoostGroupBiLevel2, but
again estimating nuisance parameters as suggested
in Liand and Zeger (also under the assumption of
only one of each)
3) hBoost is threeBoostGroupBiLevel, but
again estimating nuisance parameters as suggested
in Liand and Zeger (also under the assumption of
only one of each)
2015-07-14 Greg Vaughan
* Added implementation of third Bi-level appraoch
1) new approach selects group based on largest gamma value,
rather than largest L-2 Norm
2) implemented in threeBoostGroupBievel2.R
3) It is intended to combine all of the approach in a later update
2015-06-11 Greg Vaughan
* Fixed implementaion of Weights
1) originally, the weights were inverted before transforming the path back, but this effectively multiplies the path of estimates by the inverse TWICE
2) i.e. we want W^(-1) %*% Beta, but what was being done was W^(-2) %*% Beta