> ## 환경설정 -----
> library(tidyverse)
> 
> ## 데이터 가져오기
> dat <- read_delim("data/fundraising.txt", delim=" ")

> ## 데이터 정제
> df <- dat %>% 
+     filter(complete.cases(.)) %>% 
+     select(-no)

>     # mutate(donated = factor(donated))
> 
> ## 예측모형 적합
> df_glm <- glm(donated ~ ., data=df, family="binomial")

> ## 예측모형 변수선정
> # df_best_glm <- MASS::stepAIC(df_glm,  trace = FALSE)
> df_best_glm <- glm(donated ~ has_children + wealth_rating + pet_owner + 
+     frequency + money, data=df, family="binomial")

> ## 예측모형 회귀계수
> broom::tidy(df_best_glm)

> library(broom)
> library(extrafont)
> loadfonts()
> 
> ## Cumulative gain 그래프를 위한 데이터 준비
> 
> pred_df <- df %>% 
+     mutate(prob = predict(df_best_glm, newdata=df, type="response") ) %>% 
+     select(donated, prob, has_children, wealth_rating, pet_owner, frequency, money) %>% 
+     mutate(donated= factor(donated, levels = c(0,1)))

> # pred_df <- augment(df_best_glm, df, type.predict = "response") #.fitted
> 
> ## caret `lift`
> 
> df_lift <- caret::lift(donated ~ prob, data=pred_df, class=1, values = 50)

> ## 참조선
> lines1 <- data.frame(x1 = 0, x2 = 100, y1 = 0, y2 = 100)
> lines2 <- data.frame(x1 = 0, x2 = df_lift$pct, y1 = 0, y2 = 100)

> lines3 <- data.frame(x1 = df_lift$pct, x2 = 100, y1 = 100, y2 = 100)

> ## 시각화
> ggplot(data=df_lift$data) +
+   geom_segment(data = lines1,
+                aes(x = x1, y = y1, xend = x2, yend = y2),
+                alpha = .5, lty = 1) +
+   geom_segment(data = lines2,
+                aes(x = x1, y = y1, xend = x2, yend = y2),
+                alpha = .5, lty = 2) +
+   geom_segment(data = lines3,
+                aes(x = x1, y = y1, xend = x2, yend = y2),
+                alpha = .5, lty = 2) +
+   geom_line(aes(x = CumTestedPct, y = CumEventPct, col = liftModelVar)) +
+     theme_bw(base_family = "NanumGothic") +
+     labs(x="표본시도 백분율(%)", y="목표 획득 점유율(%)") +
+     theme(legend.position = "none") +
+   geom_vline(xintercept = df %>% summarise(mean(donated)*100) %>% pull, color="blue", linetype=3) +
+     scale_x_continuous(breaks = seq(0, 100, by = 10)) +
+     scale_y_continuous(breaks = seq(0, 100, by = 10)) +
+     geom_point(aes(x = CumTestedPct, y = CumEventPct, col = liftModelVar))

> ggplot(df_lift$data) +
+     geom_line(aes(x = 1:nrow(df_lift$data), y = lift), color = "red")  +
+     geom_point(aes(x = 1:nrow(df_lift$data), y = lift), color = "red") +
+     scale_x_continuous(breaks = seq(0, 100, by = 5)) +
+     geom_hline(yintercept = 1) +
+     scale_y_continuous(limits=c(1, 2)) +
+     theme_bw(base_family = "NanumGothic") +
+     labs(x="표본시도 백분율(%)", y="리프트(Lift)")

> library(ggplot2)
> pred_df %>% 
+     mutate(donated_num = as.integer(donated)-1L) %>% 
+     group_by(frequency) %>% 
+     summarise(size = n(),
+               incidence = mean(donated_num)) %>% 
+     ggplot(aes(x=frequency, y=size)) +
+       geom_col(width = 0.3) +
+       geom_line(aes(x=frequency, y = incidence * 1000000, group=1), size=1, color="red") +
+       geom_point(aes(x=frequency, y = incidence * 1000000, group=1), size=2, color="red")+
+       scale_y_continuous(sec.axis = sec_axis( ~. /1000000, name = "반응율"),
+                          labels = scales::comma) +
+       labs(x="기부빈도", y="기부 대상숫자") +
+       theme_bw(base_family = "NanumGothic") 

> pred_df %>% 
+     mutate(donated_num = as.integer(donated)-1L) %>% 
+     group_by(frequency) %>% 
+     summarise(size = n(),
+               incidence = mean(donated_num)) %>% 
+     DT::datatable() %>% 
+     DT::formatRound("size", digits=0) %>% 
+     DT::formatPercentage("incidence", digits=1)

> draw_pig <- function(variable) {
+     
+     # 변수명 변환
+     var_name <- enquo(variable)
+     
+     # 변수별 요약 통계량
+     tmp_df <- pred_df %>% 
+         mutate(donated_num = as.integer(donated)-1L) %>% 
+         group_by(!!var_name) %>% 
+         summarise(size = n(),
+                   incidence = mean(donated_num)) %>% 
+         ungroup()
+     
+     # 변수별 출력 요약표
+     DT_obj <- tmp_df %>% 
+         DT::datatable() %>% 
+         DT::formatRound("size", digits=0) %>% 
+         DT::formatPercentage("incidence", digits=1)
+     
+     # 변수별 출력 시각화
+     graph_obj <- tmp_df %>% 
+         ggplot(aes(x=!!var_name, y=size)) +
+           geom_bar(width = 0.3, stat ="identity") +
+           labs(y="기부 대상숫자") +
+           geom_line(aes(x=!!var_name, y = incidence * 1000000, group=1), size=1, color="red") +
+           geom_point(aes(x=!!var_name, y = incidence * 1000000, group=1), size=2, color="red")+
+  scale_y_continuous(sec.axis = sec_axis( ~. /1000000, name = "반응율"), labels = scales::comma) +
+           theme_bw(base_family = "NanumGothic") 
+ 
+     pig_list <- list(tbl = DT_obj, graph = graph_obj)
+     
+     return(pig_list)
+ }
> 
> money_list <- draw_pig(money)

> money_list$tbl

> money_list$graph

> pred_varname <- pred_df %>% colnames() %>% setdiff(c("donated", "prob"))
> 
> fund_pm_pig <- list()
> 
> for(i in seq_along(pred_varname)) {
+     cat(i, ":", pred_varname[i], "\n")
+     cat(i, ":", eval(expr(sym(pred_varname[i]))), "\n")
+     fund_pm_pig[[i]] <- draw_pig(eval(expr(sym(pred_varname[i]))))
+ }
> 
> fund_pm_pig[[1]]

> max_profit_v <- df_lift$data %>% 
+     mutate(cost = n * 15,
+            revenue = events * 300,
+            profit = revenue - cost) %>% 
+     summarise(max_profit = max(profit)) %>% 
+     pull()

> df_lift$data %>% 
+     tbl_df %>% 
+     select(-liftModelVar, -Sn, -Sp) %>% 
+     mutate(cost = n * 15,
+            revenue = events * 300,
+            profit = revenue - cost) %>% 
+     ggplot(aes(x=cost, y=profit)) +
+       geom_point() +
+       geom_line() +
+       theme_bw(base_family = "NanumGothic")  +
+       scale_x_continuous(labels = scales::comma, breaks = seq(0,1200000, 100000)) +
+       scale_y_continuous(labels = scales::comma) +
+       geom_hline(yintercept = max_profit_v, color="blue")

> df_lift$data %>% 
+     mutate(cost = n * 15,
+            revenue = events * 300,
+            profit = revenue - cost) %>% 
+     filter(between(profit, max_profit_v-100, max_profit_v+100)) %>% 
+     select(-liftModelVar, -Sn, -Sp) %>% 
+     DT::datatable() %>% 
+     DT::formatRound(c("events", "n", "cost", "revenue", "profit"), digits=0) %>% 
+     DT::formatRound(c("EventPct", "CumEventPct", "lift", "CumTestedPct"), digits=1) %>% 
+     DT::formatRound(c("cuts"), digits=3)