Plearning: Plearning

Description

This is the Plearning to estimate optimal multistage DTR. It implements improved Olearning to estimate optimal treatment rules for each stage backwardly. And it also borrows idea from Q-learning to utilize the estimated optimal outcomes for later stages.

Usage

Plearning(X,AA,RR,n,K,pi,pentype = "lasso",kernel ="linear",
sigma=c(0.03,0.05,0.07),clinear=2^(-2:2),m=4,e=1e-05)

Arguments

is either a matrix shared among all stages; or list of feature matrices, where feature matrices from different stages can have different dimensions.

a list of K, each element A[[i]] is the treatment assignment vector for stage i.

a list of K, each element R[[i]] is the outcome vector for stage i.

sample size

number of stages

a list of K, the i'th element is the randomization probability at stage i

pentype

the type of regression used to take residual, 'lasso' is the default, using lasso regression; 'LSE' is the ordianry least square regression. as in the function wsvm

clinear

is grid of tuning parameter for wsvm, and we use cross validation to choose the tuning parameter here.

number of folds in cross validation for Olearning_Single.

The rounding error for to compute bias in wsvm

kernel

The choice of kernel for Improved O-learning, default is 'linear', can also be 'rbf'

sigma

if 'rbf' is chosen for kernel, the grid of sigma to serach by cross validation.

Value

models

a list of models of class 'linearcl'

References

Liu, Y., Zeng, D., Wang, Y. (2014). Use of personalized Dynamic Treatment Regimes (DTRs) and Sequential Multiple Assignment Randomized Trials (SMARTs) in mental health studies. Shanghai archives of psychiatry, 26(6), 376. http://www.ncbi.nlm.nih.gov/pmc/articles/PMC4311115/

Liu et al. (2015). Under double-blinded review.

Examples

Run this code

# NOT RUN {
n_cluster=10
pinfo=10
pnoise=20
example2=make_2classification(n_cluster,pinfo,pnoise,100)
pi=list()
pi[[2]]=pi[[1]]=rep(1,100)
modelP=Plearning(example2$X,example2$A,example2$R,100,2,pi)
# }