app.R

library(shiny)
library(shinydashboard)
library(ggplot2)
library(dplyr)
library(readr)
library(tidyr)
library(cluster)
library(randomForest)
library(DT)

# データ読み込み
csv_path <- "FullGamePitchData.csv"
df <- read_csv(csv_path, locale = locale(encoding = "UTF-8")) %>%
  mutate(
    球速 = as.numeric(gsub("km/h", "", 球速)),
    座標top = as.numeric(座標top),
    座標left = as.numeric(座標left)
  )

ui <- dashboardPage(
  dashboardHeader(title = "NPB 投球分析"),
  dashboardSidebar(
    sidebarMenu(
      menuItem("分析結果", tabName = "analysis", icon = icon("chart-area")),
      menuItem("データベース", tabName = "data", icon = icon("table")),
      radioButtons("role", "分析対象:", choices = c("投手", "打者")),
      uiOutput("name_selector"),
      actionButton("go", "分析実行"),
      br(),
      selectInput("filter_date", "日付で絞り込み:", choices = c("すべて", unique(df$日付))),
      selectInput("filter_pitcher", "投手で絞り込み:", choices = c("すべて", unique(df$投手_名前))),
      selectInput("filter_batter", "打者で絞り込み:", choices = c("すべて", unique(df$打者_名前)))
    )
  ),
  dashboardBody(
    tabItems(
      tabItem(tabName = "analysis",
              fluidRow(
                box(title = "ヒートマップ", width = 6, plotOutput("heatmap")),
                box(title = "球種分布", width = 6, tableOutput("pitch_types"))
              ),
              fluidRow(
                box(title = "クラスタリング", width = 6, plotOutput("cluster_plot")),
                box(title = "次の球種予測", width = 6, textOutput("next_pitch"))
              )
      ),
      tabItem(tabName = "data",
              fluidRow(
                box(title = "全投球データ", width = 12, DTOutput("full_data"))
              )
      )
    )
  )
)

server <- function(input, output, session) {
  
  output$name_selector <- renderUI({
    choices <- if (input$role == "投手") unique(na.omit(df$投手_名前)) else unique(na.omit(df$打者_名前))
    selectInput("name", paste0(input$role, "名を選択:"), choices = choices)
  })
  
  filtered_data <- eventReactive(input$go, {
    col <- if (input$role == "投手") "投手_名前" else "打者_名前"
    df %>% filter(!!sym(col) == input$name)
  })
  
  output$heatmap <- renderPlot({
    d <- filtered_data()
    if (nrow(d) < 1) return(plot.new())
    ggplot(d, aes(x = 座標left, y = 座標top)) +
      stat_density2d(aes(fill = ..level..), geom = "polygon") +
      scale_fill_viridis_c() +
      ggtitle(paste(input$name, "のヒートマップ")) +
      scale_y_reverse() +
      theme_minimal()
  })
  
  output$pitch_types <- renderTable({
    d <- filtered_data()
    if (nrow(d) < 1) return(NULL)
    d %>% count(球種, name = "投球数")
  })
  
  output$cluster_plot <- renderPlot({
    d <- filtered_data() %>% drop_na(球速, 座標top, 座標left)
    if (nrow(d) < 3) return(plot.new())
    km <- kmeans(d[, c("球速", "座標top", "座標left")], centers = 3)
    d$cluster <- factor(km$cluster)
    ggplot(d, aes(x = 座標left, y = 座標top, color = cluster)) +
      geom_point(size = 3) +
      scale_y_reverse() +
      ggtitle(paste(input$name, "のクラスタリング")) +
      theme_minimal()
  })
  
  output$next_pitch <- renderText({
    d <- filtered_data() %>% drop_na(球速, 座標top, 座標left, 球種)
    if (nrow(d) < 5) return("データ不足")
    d$球種 <- as.factor(d$球種)
    model <- randomForest(球種 ~ 球速 + 座標top + 座標left, data = d)
    newdata <- d[nrow(d), c("球速", "座標top", "座標left")]
    pred <- predict(model, newdata)
    paste("次の球種予測:", as.character(pred))
  })
  
  filtered_table_data <- reactive({
    d <- df
    if (input$filter_date != "すべて") d <- d[d$日付 == input$filter_date, ]
    if (input$filter_pitcher != "すべて") d <- d[d$投手_名前 == input$filter_pitcher, ]
    if (input$filter_batter != "すべて") d <- d[d$打者_名前 == input$filter_batter, ]
    d
  })
  
  output$full_data <- renderDT({
    datatable(filtered_table_data(), options = list(pageLength = 25, scrollX = TRUE))
  })
}

shinyApp(ui, server)