04-model-tourney-sums.R

tourney_score_diff_sums <- 
  tourney_data %>% 
  group_by(year, winner_team, winner_seed) %>% 
  summarize(
    sum_score_diff = sum(winner_score - loser_score),
    ngames = n()
    ) %>% 
  rename(school = winner_team) %>% 
  ungroup()

stats_df2 <- 
  stats_df %>% 
  mutate(year = as.numeric(year)) %>% 
  dplyr::select(-starts_with("_NA..")) %>% 
  filter(!is.na(Overall_SRS)) %>% 
  dplyr::select_if(funs(!any(is.na(.))))

names(stats_df2) <- str_replace_all(names(stats_df2), "\\%", "pct")
names(stats_df2) <- str_replace_all(names(stats_df2), "\\s+","")
names(stats_df2) <- str_replace_all(names(stats_df2), "\\.", "")
names(stats_df2) <- str_replace_all(names(stats_df2), "^\\_","")
names(stats_df2) <- str_replace_all(names(stats_df2), "\\-","")

moddf <- 
  stats_df2 %>%
  left_join(tourney_score_diff_sums)

moddf <- 
  moddf %>% 
  dplyr::select(
    year, school, ngames, made_it, 
    Overall_SRS, Overall_SOS, 
    sum_score_diff, 
    contains("pct"))

moddf_train <- 
  moddf %>% 
  filter(
    year > 2010, year <= 2019,
    made_it == 1, !is.na(ngames)
    ) %>% 
  replace_na(list(sum_score_diff = 0))

moddf_test <- 
  moddf %>% 
  filter(year == 2021, made_it == 1)

rfmod <- ranger::ranger(sum_score_diff ~ . -ngames -year, data = moddf_train, num.trees = 500, importance = "impurity")