change around add horizons function, assuming last hosp data date is …

…present in df

_targets_eval_postprocessing.R

@@ -302,13 +302,8 @@ head_to_head_targets <- list(
   # forecast dates with sufficient wastewater for both ww model and hosp only
   # model. Then join the convergence df
   tar_target(
-    name = max_nowcast_days,
-    command = as.numeric(
-      max(all_ww_hosp_quantiles$forecast_date) -
-        max(all_ww_hosp_quantiles$date[!is.na(
-          all_ww_hosp_quantiles$calib_data
-        )])
-    )
+    name = last_hosp_data_date_map,
+    command = get_last_hosp_data_date_map(all_hosp_model_quantiles)
   ),
   tar_target(
     name = hosp_quantiles_filtered,
@@ -329,14 +324,18 @@ head_to_head_targets <- list(
           "forecast_date"
         )
       ) |>
+      dplyr::left_join(
+        last_hosp_data_date_map,
+        by = c("location", "forecast_date")
+      ) |>
       dplyr::left_join(
         epidemic_phases,
         by = c(
           "location" = "state_abbr",
           "date" = "reference_date"
         )
       ) |>
-      add_horizons(max_nowcast_days = max_nowcast_days)
+      add_horizons()
   ),
   # Do the same thing for the sampled scores, combining ww and hosp under
   # the status quo scenario, filtering to the locations and forecast dates
@@ -359,14 +358,18 @@ head_to_head_targets <- list(
           "forecast_date"
         )
       ) |>
+      dplyr::left_join(
+        last_hosp_data_date_map,
+        by = c("location", "forecast_date")
+      ) |>
       dplyr::left_join(
         epidemic_phases,
         by = c(
           "location" = "state_abbr",
           "date" = "reference_date"
         )
       ) |>
-      add_horizons(max_nowcast_days = max_nowcast_days)
+      add_horizons()
   ),
   # Repeat for the quantile-based scores
   tar_target(
@@ -387,14 +390,18 @@ head_to_head_targets <- list(
           "forecast_date"
         )
       ) |>
+      dplyr::left_join(
+        last_hosp_data_date_map,
+        by = c("location", "forecast_date")
+      ) |>
       dplyr::left_join(
         epidemic_phases,
         by = c(
           "location" = "state_abbr",
           "date" = "reference_date"
         )
       ) |>
-      add_horizons(max_nowcast_days = max_nowcast_days)
+      add_horizons()
   )
 )
 

wweval/NAMESPACE

@@ -27,6 +27,7 @@ export(get_hosp_values)
 export(get_input_hosp_data)
 export(get_input_ww_data)
 export(get_last_hosp_data_date)
+export(get_last_hosp_data_date_map)
 export(get_model_draws_w_data)
 export(get_model_path)
 export(get_n_states_improved_plot)

wweval/R/add_horizons.R

@@ -2,31 +2,27 @@
 #'
 #' @description
 #' This function takes in a tibble of scores on the nowcasts/forecasts
-#' containing the columns `date` and `forecast_date` and adds the following
-#' columns: `horizon_days` (an integer) and `horizon` (a string to be used for
-#' categorical grouping of horizons)
+#' containing the columns `date`, `forecast_date`, and `last_hosp_data_date`,
+#' and adds the following columns: `horizon_days` (an integer) and
+#'  `horizon` (a string to be used for categorical grouping of horizons)
 #'
 #'
 #' @param df A tibble containing either forecasts or scores (or both)
-#' and the following required columns: `date`and `forecast_date`
-#' @param max_nowcast_days An integer indicating the expected maximum number
-#' of weeks where date < forecast date. If the output contains values beyond
-#' this threshold, an error will be thrown which may indicate that the scores
-#' contain scores for calibrated data.
+#' and the following required columns: `date`,`forecast_date`,
+#' `last_hosp_data_date`
 #'
 #' @return a tibble containing the same columns as `df` plus
 #' `horizon_days` and `horizon`
 #' @export
-add_horizons <- function(df,
-                         max_nowcast_days) {
+add_horizons <- function(df) {
   df_w_horizons <- df |>
     dplyr::mutate(
       horizon_days = as.numeric(date - forecast_date)
     ) |>
     dplyr::mutate(
       horizon = dplyr::case_when(
-        horizon_days <= -!!max_nowcast_days ~ "calibration",
-        horizon_days > -!!max_nowcast_days & horizon_days <= 0 ~ "nowcast",
+        date <= last_hosp_data_date & horizon_days <= 0 ~ "calibration",
+        date > last_hosp_data_date & horizon_days <= 0 ~ "nowcast",
         horizon_days > 0 & horizon_days <= 7 ~ "1 wk",
         horizon_days > 7 & horizon_days <= 14 ~ "2 wks",
         horizon_days > 14 & horizon_days <= 21 ~ "3 wks",
@@ -37,3 +33,25 @@ add_horizons <- function(df,
 
   return(df_w_horizons)
 }
+
+#' Get a map of the location, forecast date, and last hospital admissions data
+#' date
+#'
+#' @param df A tibble containing the following columns: `forecast_date`,
+#' `location`,`date`, `calib_data`
+#'
+#' @return A tibble with that maps the unique combinations of `location` and
+#' `forecast` date to the last hospital admissions data date
+#' `last_hosp_data_date`
+#' @export
+get_last_hosp_data_date_map <- function(df) {
+  map <- df |>
+    dplyr::group_by(forecast_date, location) |>
+    dplyr::filter(!is.na(calib_data)) |>
+    dplyr::summarise(
+      last_hosp_data_date = max(date)
+    ) |>
+    dplyr::ungroup()
+
+  return(map)
+}