Chicago Suburbs Housing Data

chiburbs.R

library(tidymodels)
library(tidyverse)
library(stringr)
library(janitor)
library(doMC)
registerDoMC(cores = max(1, parallelly::availableCores() - 1))

# data cleaning --------
# we'd likely just do all this cleaning under the hood and supply 
# the `chiburbs` result as the "initial" dataset
chiburbs <- 
  # use read.csv rather than read_csv as every other line is totally blank
  # and read_csv doesn't know how to handle it
  bind_rows(
    read.csv("https://raw.githubusercontent.com/GeometricBison/HousePriceML/main/csv/naperville_2021-2022_2.csv") %>% mutate(city = "Naperville"),
    read.csv("https://raw.githubusercontent.com/GeometricBison/HousePriceML/main/csv/bolingbrook_2021-2022_2.csv") %>% mutate(city = "Bolingbrook")
  )%>%
  clean_names() %>%
  as_tibble() %>%
  rename_with(~gsub("x_of", "n", .x, fixed = TRUE)) %>%
  filter(!is.na(housingprice) & housingprice != "") %>%
  mutate(
    hoa_dues = gsub("$", "", hoa_dues, fixed = TRUE),
    hoa_dues = gsub("/month", "", hoa_dues, fixed = TRUE),
    hoa_dues = as.numeric(hoa_dues),
    hoa_dues = if_else(is.na(hoa_dues), 0, hoa_dues),
    housingprice = gsub("$", "", housingprice, fixed = TRUE),
    housingprice = gsub(",", "", housingprice, fixed = TRUE),
    housingprice = as.numeric(housingprice),
    housingprice = log(housingprice),
    sqft = gsub(",", "", sqft, fixed = TRUE),
    sqft = as.numeric(sqft),
    basement_sq_ft = gsub(",", "", basement_sq_ft, fixed = TRUE),
    basement_sq_ft = as.numeric(basement_sq_ft),
    tax_annual_amount = gsub("$", "", tax_annual_amount, fixed = TRUE),
    tax_annual_amount = gsub(",", "", tax_annual_amount, fixed = TRUE),
    tax_annual_amount = as.numeric(tax_annual_amount),
    basement_sq_ft = if_else(is.na(basement_sq_ft), 0, basement_sq_ft),
    n_baths_1_2 = if_else(is.na(n_baths_1_2), 0, n_baths_1_2),
    n_cars = if_else(is.na(n_cars), 0, n_cars),
    beds = if_else(is.na(beds), 0, beds),
    zip = str_sub(address, -5, -1),
    across(where(is.character), ~if_else(.x == "", NA_character_, .x)),
    across(where(is.character), as.factor)
  ) %>%
  rename(log_price = housingprice) %>%
  filter(!is.na(log_price)) %>%
  select(-list_price, -est_mo_payment, -basement, -address)


chiburbs

ggplot(chiburbs) +
  aes(x = sqft, y = log_price, col = city) +
  scale_x_sqrt() +
  geom_point()

# data splitting ---------------
set.seed(1)
chiburbs_split <- initial_split(chiburbs)
chiburbs_train <- training(chiburbs_split)
chiburbs_test <- testing(chiburbs_split)
chiburbs_folds <- vfold_cv(chiburbs_train)

# baseline model ---------------
fit_resamples(
  linear_reg(),
  log_price ~ .,
  chiburbs_folds,
  metrics = metric_set(rsq)
) %>%
  collect_metrics()

# more complex model ------------
recipe_basic <-
  recipe(log_price ~ ., chiburbs_train) %>%
  step_zv(all_predictors()) %>%
  step_normalize(all_numeric_predictors()) %>%
  step_filter_missing(all_predictors(), threshold = .2) %>%
  step_other(all_nominal_predictors()) %>%
  step_impute_mean(all_numeric_predictors()) %>%
  step_impute_mode(all_nominal_predictors()) %>%
  step_dummy(all_nominal_predictors())

spec_rf <-
  rand_forest() %>%
  set_mode("regression")

chiburbs_res <-
  fit_resamples(
    spec_rf,
    recipe_basic,
    chiburbs_folds,
    metrics = metric_set(rsq)
  )

collect_metrics(chiburbs_res)

chiburbs_fit <- fit(workflow(recipe_basic, spec_rf), chiburbs_train)
chiburbs_test <- augment(chiburbs_fit, new_data = chiburbs_test)

ggplot(chiburbs_test) +
  aes(x = log_price, y = .pred) +
  geom_point() +
  coord_obs_pred()