predict.ql {DTRlearn2}

Predict from a Fitted "ql" Object

Description

This function serves two purposes from a fitted "ql" object. It can recommend treatments for a new independent sample with partial or full subject features observed up to a certain stage. If subject features, treatment assignments and outcomes are fully observed in the new sample, this function can also evaluate the fitted DTR on this new sample, returning the empirical value function and benefit function.

Usage

## S3 method for class 'ql'
predict(object, H, AA=NULL, RR=NULL, K, pi=NULL, Qopt=FALSE, Qfit=FALSE, ...)

Arguments

object

fitted "ql" object

H

subject history information before treatment at the K stages for all subjects in the new sample. It should be constructed the same way as the H in fitting the ql object. See ql for how to construct H. Partial history information is allowed - when first j (j<=K) stages of H is specified, the first j stage treatments will be recommended.

AA

observed treatment assignments at the K stages for all subjects in the new sample. It is a vector if K=1, or a list of K vectors corresponding to the K stages. If not specified, treatments will be recommended for the new sample instead of DTR evaluation. The default is AA=NULL.

RR

observed outcomes at the K stages for all subjects in the new sample. It is a vector if K=1, or a list of K vectors corresponding to the K stages. If not specified, treatments will be recommended for the new sample instead of DTR evaluation. The default is RR=NULL.

K

number of stages of H observed in the new sample

pi

treatment assignment probabilities of the observed treatments at the K stages for all subjects in the new sample. It is a vector if K=1, or a list of K vectors corresponding to the K stages. It can be unspecified if one is only interested in treatment recommendations for the new sample. If both AA and RR are specified while pi is not specified, we will estimate the treatment assignment probabilities based on lasso-penalized logistic regressions with predictors being H_k at each stage k. The default is pi=NULL.

Qopt

whether to output the predicted optimal Q-function. The default is Qfun=FALSE.

Qfit

whether to output the estimated Q-function under the observed treatments. The default is fitted=FALSE.

...

further arguments passed to or from other methods.

Value

treatment

recommennded optimal treatments at the K stages for each subject in the new sample. It is a list of K vectors. If no tailoring variables are learned in the "ql" object, treatments will be assigned randomly with equal probability.

valuefun

overall empirical value function under the fitted DTR evaluated on the new sample. It is returned only when AA and RR are fully specified for the K stages.

benefit

overall empirical benefit function under the estimated DTR evaluated on the new sample. It is returned only when AA and RR are fully specified for the K stages.

pi

treatment assignment probabilities of the assigned treatments at the K stages for each subject in the new sample. If pi is not specified but H and AA are specified for the K stages, the estimated treatment assignment probabilities from lasso-penalized logistic regressions with H_k being the predictors at each stage k will be returned.

Q

the predicted optimal Q-function if Qfun=TRUE.

fitted

the estimated Q-function under the observed treatment if fitted=TRUE.

Author(s)

Yuan Chen, Ying Liu, Donglin Zeng, Yuanjia Wang

Maintainer: Yuan Chen <[e-mail]><[e-mail]>

See Also

ql, sim_Kstage, owl

Examples

# simulate 2-stage training and test sets
n_train = 100
n_test = 500
n_cluster = 10
pinfo = 10
pnoise = 20

train = sim_Kstage(n_train, n_cluster, pinfo, pnoise, K=2)
H1_train = scale(train$X)
H2_train = scale(cbind(H1_train, train$A[[1]], H1_train * train$A[[1]]))
pi_train = list(rep(0.5, n_train), rep(0.5, n_train))

test = sim_Kstage(n_test, n_cluster, pinfo, pnoise, train$centroids, K=2)
H1_test = scale(test$X)
H2_test = scale(cbind(H1_test, test$A[[1]], H1_test * train$A[[1]]))
pi_test = list(rep(0.5, n_test), rep(0.5, n_test))

# estimate DTR with ql on the training sample
ql_train = ql(H=list(H1_train, H2_train), AA=train$A, RR=train$R, K=2, pi=pi_train, m=3)

# evaluate the DTR when full information are observed on the new sample
ql_test = predict(ql_train, H=list(H1_test, H2_test), AA=test$A, RR=test$R, K=2)
ql_test$treatment
ql_test$valuefun
ql_test$pi

# recommned the first-stage treatments only
ql_test2 = predict(ql_train, H=H1_test, K=1)
ql_test2$treatment