In systats/lolR: Data Sourcing

Packages

pacman::p_load(dplyr, ggplot2, purrr, xml2, stringr, anytime, magrittr, h2o, keras, caret, slam, data.table)

Load Data

load("data/final.Rdata")
final %>% glimpse()

final_unnest <- 
  final %>% 
  tidyr::unnest()

final_sum <- final_unnest %>% 
  select(game_id, blue_longest_time_spent_living, red_longest_time_spent_living, blue_largest_critical_strike, red_largest_critical_strike, blue_total_heal, red_total_heal, blue_total_minions_killed, red_total_minions_killed, blue_wards_placed, red_wards_placed, blue_wards_killed, red_wards_killed) %>% 
  group_by(game_id) %>% 
  summarise_all(.funs = mean)

Prep

work <- final %>% 
  select(blue_win, blue_team, red_team, tournament, gp_date, game_id,
         blue_first_tower, blue_first_blood
         #blue_gold, red_gold
  ) %>%  
  left_join(final_sum) %>% 
  mutate(
    blue_win = ifelse(blue_win == "Win", 1, 0) %>% as.factor(),
    blue_team = as.factor(blue_team),
    red_team = as.factor(red_team),
    tournament = as.factor(tournament)
    #red_win = ifelse(win == "red", 1, 0)
  ) %>% 
  mutate_if(is.logical, function(x) ifelse(x, 1, 0)) 
#bind_cols(blue_champion_dummies, red_champion_dummies)

work %>% glimpse

# final %>% 
#   count(blue_win)

y <- "blue_win"
x <- setdiff(colnames(work), c(y, "gp_date", "game_id", "tournament"))

Exploration

Find Team

work %>% 
  count(blue_team)

work %>% 
  filter(game_id)


# work %>% 
#   filter(blue_team %in% c("Snake Esports", "Invictus Gaming") | red_team %in% c("Snake Esports", "Invictus Gaming"))
# 
# work %>% 
#   filter(blue_team %in% c("Royal Never Give Up", "EDward Gaming") | red_team %in% c("Royal Never Give Up", "EDward Gaming"))

Win Frequency by Team

final %>% 
  select(blue_team, red_team, blue_win) %>% 
  tidyr::gather(side, team, -blue_win) %>% 
  mutate(win = ifelse((blue_win == 1 & side == "blue_team") | (blue_win == 0 & side == "red_team"), 1, 0)) %>% 
  group_by(team) %>% 
  count(win) %>% 
  mutate(perc = n/sum(n)*100) %>% 
  filter(win == 1) %>%
  ungroup() %>% 
  filter(n > 10) %>%
  arrange(desc(perc))

By Tournament

tournament_places <- final %>% 
  select(blue_team, red_team, blue_win, tournament) %>% 
  tidyr::gather(side, team, -blue_win, -tournament) %>% 
  mutate(win = ifelse((blue_win == 1 & side == "blue_team") | (blue_win == 0 & side == "red_team"), 1, 0)) %>% 
  group_by(tournament, team) %>% 
  count(win) %>% 
  mutate(perc = n/sum(n)*100) %>% 
  filter(win == 1) %>%
  group_by(tournament) %>% 
  arrange(tournament, desc(perc)) %>% 
  mutate(place = 1:n()) %>% 
  ungroup()

tournament_places %>% 
  group_by(team) %>% 
  count(place) %>% 
  arrange(place, desc(nn))

Gold

nn <- final %>% 
  filter(!is.na(blue_gold)) %>%
  select(blue_gold, red_gold, blue_win) %>% 
  tidyr::gather(side, gold, -blue_win) 


nn %>% 
  mutate(win = ifelse((blue_win == 1 & side == "blue_gold") | (blue_win == 0 & side == "red_gold"), 1, 0) %>% as.factor) %>% 
  ggplot(aes(win, gold)) +
  geom_violin() +
  ylim(0, 200)

H2O

h2o.init()

# split_data <- function(data, p = .8){
#   train_id <- caret::createDataPartition(y = 1:nrow(data), p = p, 
#     list = F)
#   train <- data[train_id, ]
#   test <- data[-train_id, ]
#   return(list(train = train, test = test))
# } 
#train_split <- split_data(final)

train_split <- work %>% 
  split(ifelse(str_detect(.$gp_date, "2018"), "test", "train"))

train_h2o <- as.h2o(train_split$train)
#train_h2o$blue_win <- as.factor(train_h2o$blue_win)
test_h2o <- as.h2o(train_split$test)

GBM

dim(train_h2o) 
fit_gbm <- h2o.gbm(x = x, y = y, training_frame = train_h2o)
h2o.performance(fit_gbm, newdata = test_h2o)
h2o.varimp_plot(fit_gbm)

DNN

fit_dnn<- h2o.deeplearning(x = x, y = y, training_frame = train_h2o, hidden = c(200, 100, 30))
h2o.performance(fit_dnn, newdata = test_h2o)
h2o.varimp_plot(fit_dnn)

preds <- predict(fit_dnn, test_h2o) %>% 
  as_tibble %>% 
  mutate_if(is.numeric, function(x) round(x, 4))

# tibble(
#   real = train_split$test$blue_win,
#   pred = preds$predict
# )

options(scipen = 999)

final_preds <- preds %>% 
  bind_cols(train_split$test) %>%
  select(gp_date, everything()) %>%
  filter(str_detect(tournament, "MSI"))

gg1 <- final_preds %>% 
  mutate(hit = blue_win == predict) %>% 
  ggplot(aes(p1, fill = hit)) +
  geom_density(alpha = .5)

gg1


gg2 <- final_preds %>% 
  mutate(hit = blue_win == predict) %>% 
  ggplot(aes(p1, fill = hit)) +
  geom_histogram()

gg2

caret::confusionMatrix(train_split$test$blue_win,
preds$predict)

nn <- gamepedia %>% 
  select(players) %>% 
  mutate(id = 1:n()) %>%
  tidyr::unnest()

Red Champion dummy Matrix

red_champion_dummies <- nn$red_champions %>%
  as.factor() %>% 
  as.numeric() %>%
  dummies::dummy()

red_champion_dummies <- red_champion_dummies %>% 
  as_data_frame() %>%
  mutate(id = nn$id) %>% 
  group_by(id) %>% 
  summarise_all(sum) %>% 
  ungroup

Red Champion dummy Matrix

blue_champion_dummies <- nn$blue_champions %>%
  as.factor() %>% 
  as.numeric() %>%
  dummies::dummy()

blue_champion_dummies <- blue_champion_dummies %>% 
  as_data_frame() %>%
  mutate(id = nn$id) %>% 
  group_by(id) %>% 
  summarise_all(sum) %>% 
  ungroup

Red Player dummy Matrix

red_player_dummies <- nn$red_players %>%
  as.factor() %>% 
  as.numeric() %>%
  dummies::dummy()

red_player_dummies <- red_player_dummies %>% 
  as_data_frame() %>%
  mutate(id = nn$id) %>% 
  group_by(id) %>% 
  summarise_all(sum) %>% 
  ungroup

Blue Player dummy Matrix

blue_player_dummies <- nn$blue_players %>%
  dummies::dummy()

blue_player_dummies <- blue_player_dummies %>% 
  as_data_frame() %>%
  mutate(id = nn$id) %>% 
  group_by(id) %>% 
  summarise_all(sum) %>% 
  ungroup

systats/lolR documentation built on May 31, 2019, 6:17 p.m.