Computes posterior summaries of average prognostic or treatment effects based on the fitted BCF model.
average_tree_fits(fit,
type = c("mu","tau"),
X_cont = matrix(0, nrow = 1, ncol = 1),
X_cat = matrix(0, nrow = 1, ncol = 1),
quantiles = c(0.05, 0.95),
weights = NULL,
verbose = TRUE,
print_every = floor(max(c(nrow(X_cont), nrow(X_cat)))/10))A list containing
Posterior mean estimate of the desired average treatment (or prognostic) effect
Posterior quantiles of the desired average treatment (or prognostic) effect.
Object returned by flexBCF.
Character which is equal to either "mu" or "tau". Determines which tree ensemble is used to make predictions before averaging.
Matrix of continuous covariates for either the prognostic function \(\mu\) (type == "mu") or treatment effect function \(\tau\) (type == "tau") for the individuals in the group over which you wish to average. Note, predictors must be re-scaled to lie in the interval [-1,1]. Assumes that subjects are ordered so that all control subjects are listed before all treated subjects. Default is a 1x1 matrix, which signals that there are no continuous covariates
Integer matrix of categorical covariates for either the prognostic function \(\mu\) if (type == "mu") or treatment effect function \(\tau\) (type == "tau") for the individuals in the group over which you wish to average. Note categorical levels should be 0-indexed. That is, if a categorical predictor has 10 levels, the values should run from 0 to 9. Assumes that subjects are ordered so that all control subjects are listed before all treated subjects. Default is a 1x1 matrix, which signals that there are no categorical covariates
Vector of quantiles which you wish to compute. Default is c(0.05, 0.95) so that the function will return 90% posterior credible intervals for the average effect.
Vector of non-negative weights for each individual. Default is a NULL, which internally gets converted to a vector of all 1's.
Logical, inciating whether to print progress to R console. Default is TRUE.
As the MCMC runs, a message is printed every print_every iterations. Default is floor(max(c(nrow(X_cont), nrow(X_cat)))/10) so that only 10 messages are printed.
Returns posterior mean and quantiles for the conditional average treatment or prognostic effects averaged over the covariate values contained in X_cont and X_cat.
Use the argument weights to pass observation weights, which may be useful to extrapolate findings to a different population than the one observed at training time.