closes #83

R/aggregate_augmented_data_by_adm.R

@@ -9,19 +9,27 @@
 #' @author Emma Mendelsohn
 #' @export
 aggregate_augmented_data_by_adm <- function(augmented_data, 
-                                  rsa_polygon, 
-                                  model_dates_selected) {
+                                            rsa_polygon, 
+                                            model_dates_selected) {
 
+  # Read augmented data, convert to raster
   r <- arrow::read_parquet(glue::glue("{augmented_data}/date={model_dates_selected}/part-0.parquet")) |> 
     rast() 
   crs(r) <- crs(rast())
+
+  # Mask raster to polygon
   r <- mask(r, rsa_polygon) 
 
-  p <- terra::extract(r, rsa_polygon, mean, na.rm = TRUE, weights = TRUE) # weighted mean
+  # Get the mean value by polygons
+  p <- terra::extract(r, rsa_polygon, mean, na.rm = TRUE, weights = FALSE, ID = FALSE) 
+
+
+  # spatially weight in fitting - probability per unit area - per size of the district
+  # probability of it happening in the district is the spatial weighting
 
   bind_cols(rsa_polygon, p) |> 
     as_tibble() |> 
     mutate(date = model_dates_selected) |> 
-    select(date, shapeName, ID, names(p))
-
+    select(date, shapeName, names(p))
+  
 }

_targets.R

@@ -519,29 +519,47 @@ model_targets <- tar_plan(
                                              rsa_polygon, 
                                              model_dates_selected),
              pattern = model_dates_selected,
-             cue = tar_cue(tar_cue_general)
+             cue = tar_cue("thorough")
   ),
 
+  tar_target(rsa_polygon_spatial_weights, rsa_polygon |> 
+               mutate(area = st_area(rsa_polygon)) |> 
+               as_tibble() |> 
+               select(shapeName, area)),
+
   tar_target(model_data,
              left_join(aggregated_data_rsa, 
                        rvf_outbreaks, 
                        by = join_by(date, shapeName)) |>  
-               mutate(outbreak_30 = replace_na(outbreak_30, FALSE))
+               mutate(outbreak_30 = replace_na(outbreak_30, FALSE)) |> 
+               left_join(rsa_polygon_spatial_weights, by = "shapeName")
   ),
 
-  # Train/test --------------------------------------------------
-
-  # Training and Testing (holdout dataset)
-  tar_target(model_data_split, initial_split(model_data, prop = 0.8, strata = outbreak_30)),
+  # Splitting --------------------------------------------------
+  # Initial train and test (ie holdout)
+  tar_target(model_data_split, initial_split(model_data, prop = 0.8)), # pick random days left out of the training
   tar_target(training_data, training(model_data_split)),
-  tar_target(training_splits, vfold_cv(training_data, strata = outbreak_30)), # subsplit training analysis/assessment
   tar_target(holdout_data, testing(model_data_split)),
+
+  # CV splits
+  # Mask from the training set the three months following the holdout dates for the given district and the surrounding districts. 
+  # Should this be on the whole training set, or just the analysis portion of the training set?
+  # In other words, it doesn't have to be masked from the model's assessments in the CV routine, we can use that data to assess performance
+  tar_target(masked_data, get_masks_for_training(holdout_data)),
+  tar_target(training_splits, vfold_cv(training_data)), # subsplit training analysis/assessment
+  tar_target(training_splits_masked, 1), # subsplit training analysis/assessment
+
+  # Define formula and model
   tar_target(model_workflow, build_workflow(training_data)),
+
+  # Run tuning
   tar_target(tuned_parameters, tune_parameters(model_workflow, 
                                                training_splits, 
                                                grid_size = 10, 
                                                n_cores = 4))
-
+  # Fit final model
+  # Generate predictions
+
 )
 
 # Deploy -----------------------------------------------------------