Using mvsusieR for prediction

[psoerensen]

Hi everyone,
I am interested in evaluating mvsusieR for prediction purposes.
In particular I would like to compute Yhat for a test data set based on beta's estimated from a training data set.
See my example code below.
I cannot find any examples doing that or have I missed something?
(I have looked at the examples here: https://stephenslab.github.io/mvsusieR/articles/mvsusie_intro.html)
Kind regards
Peter

Simulated data from mvsusier package

data(simdata)

X <- simdata$raw$X
Y <- simdata$raw$Y

train <- 1:500

prior <- create_mixture_prior(list(matrices = simdata$par$U,
weights = simdata$par$w),
null_weight = 0)

fit <- mvsusie(X[train,],Y[train,],standardize = TRUE,
prior_variance = prior,
residual_variance = simdata$par$V,
estimate_prior_variance = TRUE,
tol = 0.01)

Extract Bhat and compute Yhat

Bhat <- fit$Bhat
Yhat <- X[-train,]%*%Bhat

[gaow]

We previously had an example for prediction ... @pcarbo should we bring that back or we should suggest people to look into eg mr.mash instead?

[pcarbo]

@gaow susieR has both a coef and predict method and I was only suggesting we do the same in mvsusieR.

[pcarbo]

@psoerensen @fmorgante I have implemented a predict method. See here for examples.

Please let us know if you have questions.

[psoerensen]

fantastic - many thanks! We will test this in the coming weeks