README.md

In jucheng1992/ctmlejl: ctmlejl: An interface to use TargetedLearning package in julia

ctmlejl: an interface to julia package TargetedLearning.jl

This package built an interfact to julia package TargetedLearning (https://lendle.github.io/TargetedLearning.jl/) based on rjulia project (https://github.com/armgong/RJulia).

For details of Targeted Learning, see the book Van der Laan, Mark J., and Sherri Rose. Targeted learning: causal inference for observational and experimental data. Springer Science & Business Media, 2011., or https://lendle.github.io/TargetedLearning.jl/

Installation julia

Before using this package, ensure you have julia (v0.4.x.) installed. If not, you can download latest version of julia in http://julialang.org/downloads/

For mac users, use symbolic link so you can simply call julia in terminal (else rjulia package can not be installed):

ln -s /Applications/Julia-0.4.1.app/Contents/Resources/julia/bin/julia /usr/local/bin/julia

Finally, simply open julia by type julia in terminal. In julia, you need install several packages:

Pkg.add("GLM")
Pkg.clone("https://github.com/lendle/TargetedLearning.jl.git")

Installation

Install devtools package:

install.packages("devtools")

Install rjulia package from github:

devtools::install_github("armgong/rjulia", ref="master")

Install ctmlejl package from github:

devtools::install_github("jucheng1992/ctmlejl")

Initialize julia

For some technical reasons, this package is not able to automatic load julia package for you. You may manually initialize it:

julia_init()
julia_void_eval("using DataArrays")
julia_void_eval("using NumericExtensions")
julia_void_eval("using Distributions")
julia_void_eval("using TargetedLearning")
julia_void_eval("using DataFrames")
julia_void_eval("using GLM")
julia_void_eval("import StatsBase.predict")

Examples using cTMLE model:

Here is some examples about using cTMLE models:

logit <- function(x){
      res = log(x/ (1-x))
      return(res)
}

n <- 1000
p <- 10
QnA1 <- runif(n)
QnA0 <- runif(n)
w <- matrix(rnorm(n * p), n, p)
a <- sample(c(0,1), n, replace = TRUE)
y <- sample(c(0,1), n, replace = TRUE)
gn1 <- runif(n)

tmle(logit(QnA1), logit(QnA0), w, a, y, gn1,
     param = "ATE")

ctmle(logit(QnA1), logit(QnA0), w, a, y, v=5, gbounds=c(0.025,0.975), patience =p, param = "Mean0")
ctmle(logit(QnA1), logit(QnA0), w, a, y, v=5, gbounds=c(0.025,0.975), patience =p, param = "Mean1")
ctmle(logit(QnA1), logit(QnA0), w, a, y, v=5, gbounds=c(0.025,0.975), patience =p, param = "ATE")


ctmle(logit(QnA1), logit(QnA0), w, a, y, v=5, gbounds=c(0.025,0.975),
      patience =p, searchstrategy = "LogisticOrdering")

ctmle(logit(QnA1), logit(QnA0), w, a, y, v=5, gbounds=c(0.025,0.975),
      patience =p, searchstrategy = "ManualOrdering", order = 2:11)

jucheng1992/ctmlejl documentation built on May 20, 2019, 4:20 a.m.