R: Drawing Posterior Predictive Realizations for NFT BART...

predict.nft2 {nftbart}

R Documentation

Drawing Posterior Predictive Realizations for NFT BART models.

Description

The function predict.nft2()/predict.nft() is the main function for drawing posterior predictive realizations at new inputs using a fitted model stored in a nft2/nft object returned from nft2()/nft().

Usage

## S3 method for class 'nft2'
predict(
            ## data
            object,
            xftest=object$xftrain,
            xstest=object$xstrain,
            ## multi-threading
            tc=getOption("mc.cores", 1), ##OpenMP thread count
            ## current process fit vs. previous process fit
            XPtr=TRUE,
            ## predictions
            K=0,
            events=object$events,
            FPD=FALSE,
            probs=c(0.025, 0.975),
            take.logs=TRUE,
            na.rm=FALSE,
            RMST.max=NULL,
            ## default settings for NFT:BART/HBART/DPM
            fmu=object$NFT$fmu,
            soffset=object$soffset,
            drawDPM=object$drawDPM,
            ## etc.
            ...)

## S3 method for class 'nft'
predict(
            ## data
            object,
            x.test=object$x.train,
            ## multi-threading
            tc=getOption("mc.cores", 1), ##OpenMP thread count
            ## current process fit vs. previous process fit
            XPtr=TRUE,
            ## predictions
            K=0,
            events=object$events,
            FPD=FALSE,
            probs=c(0.025, 0.975),
            take.logs=TRUE,
            na.rm=FALSE,
            RMST.max=NULL,
            ## default settings for NFT:BART/HBART/DPM
            fmu=object$NFT$fmu,
            soffset=object$soffset,
            drawDPM=object$drawDPM,
            ## etc.
            ...)

Arguments

`object`	Object of type `nft2/nft` from a previous call to `nft2()/nft()`.
`xftest`, `xstest`, `x.test`	New input settings in the form of a matrix at which to construct predictions. Defaults to the training inputs.
`tc`	Number of OpenMP threads to use for parallel computing.
`XPtr`	If `object` was created during the currently running R process, then (via an Rcpp `XPtr`) the BART/HBART tree ensemble objects can be accessed in RAM; otherwise, those objects will need to be loaded from their string encodings.
`K`	The length of the grid of time-points to be used for survival predictions. Set to zero to avoid these calculations which can be time-consuming for large data sets.
`events`	You can specify the grid of time-points; otherwise, they are derived from quantiles of the augmented event times.
`FPD`	Whether to yield the usual predictions or marginal predictions calculated by the partial dependence function.
`probs`	A vector of length two containing the lower and upper quantiles to be calculated for the predictions.
`take.logs`	Whether or not to take logarithms.
`na.rm`	Whether `NA` values should be removed from the summaries.
`RMST.max`	To calculate Restricted Mean Survival Time (RMST), we need to set a reasonable time maxima. Typically, a clinically important time that a majority (or a large plurality) of censored subjects have been followed through that point or beyond.
`fmu`	BART centering parameter for the test data. Defaults to the value used by `nft2()/nft()` when training the model.
`soffset`	HBART centering parameter for the test data. Defaults to the value used by `nft2()/nft()` when training the model.
`drawDPM`	Whether NFT BART was fit with, or without, DPM.
`...`	The et cetera objects passed to the `predict` method. Currently, it has no functionality.

Details

predict.nft2()/predict.nft() is the main function for calculating posterior predictions and uncertainties once a model has been fit by nft2()/nft().

Returns a list with the following entries.

Value

`f.test`	Posterior realizations of the mean function stored in a matrix. Omitted if partial dependence functions are performed since these will typically be large.
`s.test`	Posterior realizations of the SD function stored in a matrix. Omitted if partial dependence functions are performed since these will typically be large.
`f.test.mean`	Posterior predictive mean of mean function.
`f.test.lower`	Posterior predictive lower quantile of mean function.
`f.test.upper`	Posterior predictive upper quantile of mean function.
`s.test.mean`	Posterior predictive mean of SD function.
`s.test.lower`	Posterior predictive lower quantile of SD function.
`s.test.upper`	Posterior predictive upper quantile of SD function.
`surv.fpd`	Survival function posterior draws on a grid of time-points by the partial dependence function when requested.
`surv.fpd.mean`	Survival function estimates on a grid of time-points by the partial dependence function when requested.
`surv.fpd.lower`	Survival function lower quantiles on a grid of time-points by the partial dependence function when requested.
`surv.fpd.upper`	Survival function upper quantiles on a grid of time-points by the partial dependence function when requested.

Author(s)

Rodney Sparapani: rsparapa@mcw.edu