README DOCUMENT FOR THE sss PACKAGE

AUTHOR: BRIAN M. BOT

ATTRIBUTION: Hans C, Dobra A, and West M. Shotgun Stochastic Search for "Large p" Regression. Journal of the American Statistical Association 2007.

Wang, Q. for the source code

This current version uses pre-compiled binaries. If one's system architecture is not represented by one of these, the package will not build.
A new version which will allow dynamic compilation is forthcoming.

#######################################################################

# sssVignette.R

# AUTHORS
# Erich S. Huang & Brian Bot
# Sage Bionetworks
# Seattle, Washington

# SOURCE
# https://github.com/Sage-Bionetworks/sss

# NOTES
# A vignette for using the new "sss" package which provides R integration for
# Shotgun Stochastic Search by Chris Hans, Adrian Dobra and Mike West

# INSTALL SSS
# Alternative #1
# Download from Github at https://github.com/Sage-Bionetworks/sss and 
# "R CMD INSTALL sss" from the parent directory

# Alternative #2
# Install "devtools" from CRAN
require(devtools)
install_github("sss", "Sage-Bionetworks")

## START MODELING
require(sss)
require(Biobase) # from Bioconductor
require(breastCancerTRANSBIG) # from Bioconductor

data(transbig)
expressData <- exprs(transbig)[1:500, ] # Subset the data for convenience
pheno <- phenoData(transbig)

# CREATE TRAINING AND VALIDATION COHORTS
set.seed(031512)
randVec <- rbinom(dim(transbig)[2], size = 1, prob = 0.5)
trainExpress <- expressData[ , randVec == 0]
validExpress <- expressData[ , randVec == 1]
trainScore <- as.numeric(pheno@data$er[randVec == 0])
validScore <- as.numeric(pheno@data$er[randVec == 1])

# BINARY MODEL OF 'ER Status' USING SSS
sssERFit <- sss(trainScore ~ t(trainExpress))
print(sssERFit)

# print output below:
# Type of model fit: "sssBinaryModel"
# 
# Call: sss(formula = trainScore ~ t(trainExpress))
# Number of features searched : 500
# Number of training samples  : 85
# To test this predictive model against a validation set, pass a new feature matrix to:
#   predict(object, newdata=newFeatureMatrix)
# 
# ----------------------
#   Contains slots (class)
# ----------------------
#   standScore (numeric)
# postMargProb (numeric)
# trainPredictionSummary (numeric)
# testPredictionSummary (numeric)
# model (sssBinaryModel)
# nBestFits (list)
# p
# score
# indices
# pmode
# pvar
# trainPrediction
# testPrediction

# PREDICT ON THE VALIDATION SET
validScoreHat <- predict(sssERFit, newdata = t(validExpress))