cmss {studyStrap}

Covariate-Matched Study Strap for Multi-Study Learning: Fits accept/reject algorithm based on covariate similarity measure

Description

Covariate-Matched Study Strap for Multi-Study Learning: Fits accept/reject algorithm based on covariate similarity measure

Usage

cmss(formula = Y ~ ., data, target.study, sim.fn = NA,
  converge.lim = 50000, bag.size = length(unique(data$Study)),
  max.straps = 150, paths = 5, stack = "standard", sim.covs = NA,
  ssl.method = list("lm"), ssl.tuneGrid = list(c()), sim.mets = TRUE,
  model = FALSE, meanSampling = FALSE, customFNs = list(),
  stack.standardize = FALSE)

Arguments

formula

Model formula

data

A dataframe with all the studies has the following columns in this order: "Study", "Y", "V1", ...., "Vp"

target.study

Dataframe of the design matrix (just covariates) of study one aims to make predictions on

sim.fn

Optional function to be used as similarity measure for accept/reject step. Default function is: |cor( barx^(r)|,~ barx_target ) |

converge.lim

Integer indicating the number of consecutive rejected study straps to reach convergence criteria.

bag.size

Integer indicating the bag size tuning parameter.

max.straps

Integer indicating the maximum number of accepted straps that can be fit across all paths before the algorithm stops accepting new study straps.

paths

Integer indicating the number of paths (an accept/reject path is all of the models accepted before reaching one convergence criteria).

stack

String determining whether stacking matrix made on training studies "standard" or on the accepted study straps "ss." Default: "standard."

sim.covs

Is a vector of names of covariates or the column numbers of the covariates to be used for the similarity measure. Default is to use all covariates.

ssl.method

A list of strings indicating which modeling methods to use.

ssl.tuneGrid

A list of the tuning parameters in the format of the caret package. Each element must be a dataframe (as required by caret). If no tuning parameters are required then NA is indicated.

sim.mets

Boolean indicating whether to calculate default covariate profile similarity measures.

model

Indicates whether to attach training data to model object.

meanSampling

= FALSE Boolean determining whether to use mean covariates for similarity measure. This can be much quicker.

customFNs

Optional list of functions that can be used to add custom covaraite profile similarity measures.

stack.standardize

Boolean determining whether stacking weights are standardized to sum to 1. Default is FALSE

Value

A model object of studyStrap class "ss" that can be used to make predictions.

Examples

##########################
##### Simulate Data ######
##########################

set.seed(1)
# create half of training dataset from 1 distribution
X1 <- matrix(rnorm(2000), ncol = 2) # design matrix - 2 covariates
B1 <- c(5, 10, 15) # true beta coefficients
y1 <- cbind(1, X1) %*% B1

# create 2nd half of training dataset from another distribution
X2 <- matrix(rnorm(2000, 1,2), ncol = 2) # design matrix - 2 covariates
B2 <- c(10, 5, 0) # true beta coefficients
y2 <- cbind(1, X2) %*% B2

X <- rbind(X1, X2)
y <- c(y1, y2)

study <- sample.int(10, 2000, replace = TRUE) # 10 studies
data <- data.frame( Study = study, Y = y, V1 = X[,1], V2 = X[,2] )

# create target study design matrix for covariate profile similarity weighting and
# accept/reject algorithm (covaraite-matched study strap)
target <- matrix(rnorm(1000, 3, 5), ncol = 2) # design matrix
colnames(target) <- c("V1", "V2")

##########################
##### Model Fitting #####
##########################

# Fit model with 1 Single-Study Learner (SSL): PCA Regression
arMod1 <-  cmss(formula = Y ~.,
               data = data,
               target.study = target,
               converge.lim = 10,
               bag.size = length(unique(data$Study)),
               max.straps = 50,
               paths = 2,
               ssl.method = list("pcr"),
               ssl.tuneGrid = list(data.frame("ncomp" = 2))
               )

# Fit model with 2 SSLs: Linear Regression and PCA Regression
arMod2 <-  cmss(formula = Y ~.,
               data = data,
               target.study = target,
               converge.lim = 20,
               bag.size = length(unique(data$Study)),
               max.straps = 50,
               paths = 2,
               ssl.method = list("lm", "pcr"),
               ssl.tuneGrid = list(NA, data.frame("ncomp" = 2))
               )



# Fit model with custom similarity function for
# accept/reject step and 2 custom function for Covariate
# Profile Similarity weights

# custom function for CPS

fn1 <- function(x1,x2){
return( abs( cor( colMeans(x1), colMeans(x2) )) )
}

fn2 <- function(x1,x2){
return( sum ( ( colMeans(x1) - colMeans(x2) )^2 ) )
}

arMod3 <-  cmss(formula = Y ~.,
               data = data,
               target.study = target,
               sim.fn = fn1,
               customFNs = list(fn1, fn2),
               converge.lim = 50,
               bag.size = length(unique(data$Study)),
               max.straps = 50,
               paths = 2,
               ssl.method = list("lm", "pcr"),
               ssl.tuneGrid = list(NA, data.frame("ncomp" = 2))
               )

#########################
#####  Predictions ######
#########################

preds <- studyStrap.predict(arMod1, target)