Fix final 2024 res model export issues (#209)

* Fix building-level unprorated predicted value calculation

It took me literally all day to figure out this single change for
multi-card prorated PINs

* Fix missing var and typos in comps report section

* Coerce n_years_exe to int to maintain Athena schema

* Set final 2024 res run for export

* Set final parameter values for 2024

* Update DR template notes and headings

* Swap card sheet column order

* Fix script ToC

* Skip recoding char_apts variable

* Adjust char_apts handling to match changes to ingest

* Replace NA handling for char_apts

* Update final run ID

* Fix char_apts display in workbooks

params.yaml

@@ -16,7 +16,7 @@ run_type: "test"
 # Note included with each run. Use this to summarize what changed about the run
 # or add context
 run_note: |
-  Cutover to 2024 model (Chicago reassessment) using 2023 data
+  Final 2024 residential model with updated data and new features
 
 toggle:
   # Should the train stage run full cross-validation? Otherwise, the model
@@ -25,10 +25,10 @@ toggle:
 
   # Should SHAP values be calculated for this run in the interpret stage? Can be
   # desirable to save time when testing many models
-  shap_enable: FALSE
+  shap_enable: TRUE
 
   # Should comps be calculated for this run in the interpret stage?
-  comp_enable: FALSE
+  comp_enable: TRUE
 
   # Upload all modeling artifacts and results to S3 in the upload stage. Set
   # to FALSE if you are not a CCAO employee
@@ -444,5 +444,5 @@ comp:
 # Final run ID(s) chosen for export to Desk Review spreadsheets and iasWorld
 # upload
 export:
-  triad_code: "3"
-  run_id: "2023-03-14-clever-damani"
+  triad_code: "1"
+  run_id: "2024-02-06-relaxed-tristan"

pipeline/02-assess.R

@@ -81,7 +81,7 @@ assessment_card_data_mc <- assessment_card_data_pred %>%
   # For prorated PINs with multiple cards, take the average of the card
   # (building) across PINs. This is because the same prorated building spread
   # across multiple PINs sometimes receives different values from the model
-  group_by(meta_tieback_key_pin, meta_card_num) %>%
+  group_by(meta_tieback_key_pin, meta_card_num, char_land_sf) %>%
   mutate(
     pred_card_intermediate_fmv = ifelse(
       is.na(meta_tieback_key_pin),
@@ -208,18 +208,16 @@ message("Prorating buildings")
 assessment_pin_data_prorated <- assessment_pin_data_w_land %>%
   group_by(meta_tieback_key_pin) %>%
   mutate(
-    tieback_total_land_fmv = ifelse(
+    # 1. Determine the mean, unprorated building value for buildings that span
+    # multiple PINs. This is the mean value of the predicted value minus land
+    pred_pin_final_fmv_bldg_no_prorate = ifelse(
       is.na(meta_tieback_key_pin),
-      pred_pin_final_fmv_land,
-      sum(pred_pin_final_fmv_land)
+      pred_pin_final_fmv_round_no_prorate - pred_pin_final_fmv_land,
+      mean(pred_pin_final_fmv_round_no_prorate - pred_pin_final_fmv_land)
     )
   ) %>%
   ungroup() %>%
   mutate(
-    # 1. Subtract the TOTAL value of the land of all linked PINs. This leaves
-    # only the value of the building that spans the PINs
-    pred_pin_final_fmv_bldg_no_prorate =
-      pred_pin_final_fmv_round_no_prorate - tieback_total_land_fmv,
     # 2. Multiply the building by the proration rate of each PIN/card. This is
     # the proportion of the building's value held by each PIN
     pred_pin_final_fmv_bldg =
@@ -314,18 +312,22 @@ message("Saving card-level data")
 
 # Keep only card-level variables of interest, including: ID variables (run_id,
 # pin, card), characteristics, and predictions
+char_vars <- params$model$predictor$all[
+  grepl("^char_", params$model$predictor$all)
+]
+char_vars <- char_vars[!char_vars %in% c("char_apts", "char_recent_renovation")]
 assessment_card_data_merged %>%
   select(
     meta_year, meta_pin, meta_class, meta_card_num, meta_card_pct_total_fmv,
     meta_complex_id, pred_card_initial_fmv, pred_card_final_fmv, char_class,
     all_of(params$model$predictor$all), township_code
   ) %>%
-  mutate(meta_complex_id = as.numeric(meta_complex_id)) %>%
-  ccao::vars_recode(
-    starts_with("char_"),
-    type = "long",
-    as_factor = FALSE
+  mutate(
+    meta_complex_id = as.numeric(meta_complex_id),
+    ccao_n_years_exe_homeowner = as.integer(ccao_n_years_exe_homeowner),
+    char_apts = as.character(char_apts)
   ) %>%
+  ccao::vars_recode(any_of(char_vars), type = "long", as_factor = FALSE) %>%
   write_parquet(paths$output$assessment_card$local)
 
 

pipeline/07-export.R

@@ -281,12 +281,12 @@ num_apts_by_pin <- assessment_card %>%
   # to scan in a spreadsheet
   mutate(
     char_apts = case_when(
-      char_apts == "None" | is.na(char_apts) ~ "Missing",
-      char_apts == "Two" ~ "2",
-      char_apts == "Three" ~ "3",
-      char_apts == "Four" ~ "4",
-      char_apts == "Five" ~ "5",
-      char_apts == "Six" ~ "6",
+      char_apts == "NONE" | is.na(char_apts) ~ NA_character_,
+      char_apts == "TWO" ~ "2",
+      char_apts == "THREE" ~ "3",
+      char_apts == "FOUR" ~ "4",
+      char_apts == "FIVE" ~ "5",
+      char_apts == "FIX" ~ "6",
       TRUE ~ "Missing"
     )
   ) %>%
@@ -431,12 +431,6 @@ assessment_pin_prepped <- assessment_pin_merged %>%
       property_full_address,
       "[^[:alnum:]|' ',.-]"
     ),
-    # char_apts should only apply to 211s and 212s
-    char_apts = ifelse(
-      (meta_class != "211" & meta_class != "212"),
-      NA,
-      char_apts
-    ),
     # char_ncu should only apply to 212s
     char_ncu = ifelse(meta_class != "212", NA, char_ncu)
   )
@@ -464,21 +458,13 @@ assessment_card_prepped <- assessment_card %>%
     # Make sure the format of char_apts matches the short format we used to
     # generate assessment_pin_prepped
     char_apts = case_when(
-      char_apts == "None" | is.na(char_apts) ~ "Missing",
-      char_apts == "Two" ~ "2",
-      char_apts == "Three" ~ "3",
-      char_apts == "Four" ~ "4",
-      char_apts == "Five" ~ "5",
-      char_apts == "Six" ~ "6",
+      char_apts == "NONE" | is.na(char_apts) ~ NA_character_,
+      char_apts == "TWO" ~ "2",
+      char_apts == "THREE" ~ "3",
+      char_apts == "FOUR" ~ "4",
+      char_apts == "FIVE" ~ "5",
+      char_apts == "FIX" ~ "6",
       TRUE ~ "Missing"
-    )
-  ) %>%
-  mutate(
-    # char_apts should only apply to 211s and 212s
-    char_apts = ifelse(
-      (meta_class != "211" & meta_class != "212"),
-      NA,
-      char_apts
     ),
     # char_ncu should only apply to 212s
     char_ncu = ifelse(meta_class != "212", NA, char_ncu)
@@ -508,7 +494,7 @@ for (town in unique(assessment_pin_prepped$township_code)) {
     filter(township_code == town) %>%
     select(-township_code)
 
-  # 5.3 Comp details -----------------------------------------------------------
+  ## 5.1 Comp details ----------------------------------------------------------
 
   # Filter the training data so that we only display sales that are referenced.
   # First, get the indexes of every sale whose comp is referenced in
@@ -525,7 +511,7 @@ for (town in unique(assessment_pin_prepped$township_code)) {
   training_data_selected <- training_data_filtered %>%
     select(
       meta_pin, meta_sale_price, meta_sale_date, meta_class, meta_nbhd_code,
-      char_yrblt, loc_property_address, char_beds, char_ext_wall,
+      loc_property_address, char_yrblt, char_beds, char_ext_wall,
       char_heat, char_bldg_sf, char_type_resd, char_land_sf
     ) %>%
     ccao::vars_recode(type = "long")
@@ -568,7 +554,7 @@ for (town in unique(assessment_pin_prepped$township_code)) {
     startCol = 1, startRow = 5, colNames = FALSE
   )
 
-  ## 5.2. PIN-Level ------------------------------------------------------------
+  # 5.2. PIN-Level -------------------------------------------------------------
 
   # Update PIN-level data to link to comps detail sheet
   training_data_ids <- training_data_filtered %>%
@@ -859,10 +845,9 @@ upload_data <- assessment_pin %>%
   ungroup() %>%
   select(
     township_code,
-    PARID = meta_pin, CARD = meta_card_num,
-    USER37 = pred_card_final_fmv_no_prorate,
-    USER24 = meta_tieback_proration_rate.x,
-    OVRRCNLD = pred_card_final_fmv
+    PARID = meta_pin,
+    CARD = meta_card_num,
+    MV = pred_card_final_fmv_no_prorate
   )
 
 

reports/_setup.qmd

@@ -1,6 +1,6 @@
 ---
 params:
-  run_id: "2024-01-29-boring-carly"
+  run_id: "2024-02-06-relaxed-tristan"
   year: "2024"
 ---
 

reports/performance/_comp.qmd

@@ -18,10 +18,7 @@ homogeneous properties.
 ```{r _comp_neighborhood_means_coding}
 base_data <- assessment_card %>%
   mutate(triad = ccao::town_get_triad(as.character(township_code))) %>%
-  select(
-    meta_pin, meta_nbhd_code, meta_class, triad,
-    meta_card_num, meta_modeling_group
-  ) %>%
+  select(meta_pin, meta_nbhd_code, meta_class, triad, meta_card_num) %>%
   left_join(comp_df, by = c("meta_pin" = "pin", "meta_card_num" = "card"))
 
 generate_plot <- function(df) {
@@ -60,10 +57,10 @@ plots$"Overall" <- base_data %>%
   filter(triad == run_triad_code) %>%
   generate_plot()
 plots$"Single-Family" <- base_data %>%
-  filter(meta_modeling_group == "SF", triad == run_triad_code) %>%
+  filter(!(meta_class %in% c("211", "212")), triad == run_triad_code) %>%
   generate_plot()
 plots$"Multi-Family" <- base_data %>%
-  filter(meta_modeling_group == "MF", triad == run_triad_code) %>%
+  filter(meta_class %in% c("211", "212"), triad == run_triad_code) %>%
   generate_plot()
 ```
 
@@ -227,7 +224,7 @@ leaflet() %>%
       "<hr>",
       "Certified FMV: ",
       scales::dollar(meta_certified_tot, accuracy = 1),
-      "<br>Inital Card FMV: ",
+      "<br>Initial Card FMV: ",
       scales::dollar(pred_card_initial_fmv, accuracy = 1),
       "<hr>",
       "Sale Date: ", meta_sale_date,

reports/performance/performance.qmd

@@ -20,7 +20,7 @@ knitr:
     out.width: "100%"
 editor: source
 params:
-  run_id: "2024-01-29-boring-carly"
+  run_id: "2024-02-06-relaxed-tristan"
   year: "2024"
 ---