SL.stack: SL.stack

In jlstiles/sim.papers:

Description

Function that runs fully nested SuperLearner cross validated estimates on V-folds. Only supports binary treatment.

Usage

SL.stack(Y, X, A, W, newdata, method, SL.library, SL.libraryG, V = 10,
  mc.cores = 1, ...)

Arguments

Y,	outcome vector
X,	data.frame of variables that Y is a function of.
A,	treatment vector
W,	vector of variables A is a function of.
newdata,	dataframe of X, stacked with X when A=1 and X when A=0, in that order
method,	the SuperLearner meta learning method
SL.library,	SuperLearner Library for finding outcome model
SL.libraryG,	SuperLearner Library for the treatment mechanism
V,	the number of folds
mc.cores,	number of cores to use for parallel processing the SuperLearner. Note, this parallelizes across the folds not within SuperLearner

Value

A list with 5 elements: initdata: the initdata argument for running tmle with gentmle function

Qcoef: the avg SuperLearner coef for each model in the outcome regression

Gcoef: the avg SuperLearner coef for each model in the treatment mech regression

Qrisk: the avg SuperLearner risk for each model in the outcome regression

Grisk: the avg SuperLearner risk for each model in the treatment mech

inds: the indices for all the val sets, stacked to match

Examples

# basic example with very simple SuperLearner library
# generate the data according to a couple of built-in functions here

data = gendata(1000, g0 = g0_linear, Q0 = Q0_trig1)

#drop Y
Y = data$Y
A = data$A
#drop Y to form X and form newdata
X = data[,-6]
X0 = X1 = X
X0$A = 0
X1$A = 1
newdata = rbind(X,X1,X0)
#form W
W = X[,-1]

#declare SL library
SL.library = SL.libraryG = c("SL.glm","SL.mean")

stack = SL.stack(Y, X, A, W, newdata, method = "method.NNloglik",
                 SL.library, SL.libraryG, V=10, mc.cores = 4)

# simultaneously run one-step tmle for ATE and blip variance with 
# simultaneous CI
tmle.info = gentmle(initdata=stack$initdata, 
                    params=list(param_ATE,param_sigmaATE), 
                    submodel = submodel_logit, loss = loss_loglik,
                    approach = "recursive", max_iter = 10000, 
                    g.trunc = 1e-2,
                    simultaneous.inference = TRUE)
tmle.info$steps
# get simultaneous CIs
ci_gentmle(tmle.info)

jlstiles/sim.papers documentation built on May 23, 2019, 5:03 a.m.