create meta_learner fit and predict - basic

DESCRIPTION

@@ -8,7 +8,7 @@ Authors@R:
              email = "[email protected]"))
 Description: Near Real Time air pollution model results and code produced by the SET group. It is fully tested, versioned, and open source and open access.
 Depends: R (>= 3.5.0)
-Imports: dplyr, sf, stats, methods
+Imports: dplyr, sf, stats, methods, BART
 Suggests: covr, withr, knitr, rmarkdown, testthat (>= 3.0.0), terra, stars, sftime
 Encoding: UTF-8
 VignetteBuilder: knitr, rmarkdown

NAMESPACE

@@ -8,3 +8,6 @@ export(check_means_are_valid)
 export(check_output_locations_are_valid)
 export(generic_base_learner)
 export(generic_meta_learner)
+export(meta_learner_fit)
+export(meta_learner_predict)
+export(vec2nc)

R/meta_learner.R

@@ -0,0 +1,75 @@
+#' meta_learner_fit
+#' Fit a BART (Bayesian Additive Regression Tree) meta learner. It takes
+#' predictions of other models such as kriging, GLM, machine learning models as
+#' input and fits a BART Model
+#' @param base_predictor_list - a list where each
+#' @param y dependent variable
+#' @param kfolds integer, index of k-folds for cross-validation. This should be
+#' produced with regards to spatial and/or temporal considerations
+#' @return meta_fit_obj object of meta learner
+#' @export
+#' @examples NULL
+meta_learner_fit <- function(base_predictor_list,
+                             kfolds, y) {
+
+  # check lengths of each base predictor 
+  if (sapply(base_predictor_list, length, simplify = TRUE) |> 
+      stats::var() != 0) {
+    print("WARNING: base predictors need to be the same length")
+  }
+
+  # convert list to data.frame
+  x.design <- as.data.frame(base_learner_list)
+
+  # Unique k-folds (typically 5 or 10)
+  n.k <- length(unique(kfolds))
+  # Pre-allocate list of meta objects
+  meta_fit_obj <- vector(mode = "list", length = n.k)
+  for (i in 1:n.k) {
+    # get the training and test sets
+    x.tr <- x.design[i != n.k,]
+    x.te <- x.design[i == n.k,]
+    y.tr <- y[i != n.k]
+    # Fit the BART model
+    bart.fit[[i]] <- mc.wbart(x.tr, y.tr, x.test = x.te)
+  }
+  return(meta_fit_obj)
+}
+
+
+
+#' meta_learner_predict - take the list of BART fit objects and prediction 
+#' location info to create meta_learner predictions
+#'
+#' @param meta_fit_obj list of BART objects from meta_learner_fit
+#' @param pred_loc sf or dataframe grid of prediction locations
+#' @return meta_pred_nc NetCDF (nc) file of the final meta learner predictions
+#' @export
+#'
+#' @examples NULL
+meta_learner_predict <- function(meta_fit_obj, pred_loc) {
+
+  #Use the predict method
+  meta_pred_vec <- predict(meta_fit_obj, new = pred_loc)
+  # Call the vec2nc function to create a NetCDF (nc) file from the
+  # vector predictions
+  meta_pred_nc <- vec2nc(meta_pred_vec)
+  return(meta_pred_nc)
+
+}
+
+
+#' vec2nc takes a vector or matrix of predictions, some spatial information,
+#' and create a NetCDF
+#' @param pred_vec vector/matrix of predictions
+#' @param pred_bbox an sf bounding box defining the domain of the NetCDF
+#' @param field_names vector of field names for each layer in the NetCDF
+#' @return nc_out a NetCDF file
+#' @export
+#'
+#' @examples NULL
+vec2nc <- function(pred_vec, pred_bbox, field_names) {
+  nc_out <- pred_vec + pred_bbox
+  names(nc_out) <- field_names
+  return(nc_out)
+}