闪亮的Flexdasboard反应式压力计未更新

问题描述 投票:1回答:1

我已在我的Shinyapp中添加了一个反应计。量表应显示与运动员先前所有时间的最低和最高值相比最近的跳跃高度得分。

selectInput被设置为Athlete和最近的日期(max(jumpdata$Date))。我的代码对于无功仪表的最大值非常有效,但不会在无功更新最小值。当我运行该应用程序时,最小值将显示第一个运动员的输入,然后在我更新并选择其他输入(但最大更改)时保持在相同的值。

我不确定障碍物在哪里,因为最大值正在更新。

ui.r

library(shiny)
library(shinydashboard)
library(flexdashboard)
library(dplyr)

jumpdata <- read.csv("SO CMJ Dummy.csv")
jumpdata$Date <- as.Date(jumpdata$Date, "%Y-%m-%d")

shinyUI(
    fluidPage(
 sidebarPanel(width = 3,
         selectInput("Athlete", label = "Athlete",
                     choices = unique(jumpdata$Athlete))),
       mainPanel(
        fluidRow(
            box(title = "Jump Height", gaugeOutput("Gauge_JH"))
                            ))
                   ))

server.r

library(shiny)
library(shinydashboard)
library(flexdashboard)
library(dplyr)

jumpdata <- read.csv("SO CMJ Dummy.csv")
jumpdata$Date <- as.Date(jumpdata$Date, "%Y-%m-%d")

shinyServer(function(input, output){
    output$Gauge_JH <- renderGauge({
        f <- jumpdata %>%
            select(Date, Athlete, JumpHeight_cm) %>%
            filter(Athlete == input$Athlete & Date == c(max(jumpdata$Date)))

        t <- jumpdata %>%
            select(Date, Athlete, JumpHeight_cm) %>%
            filter(Athlete == input$Athlete)

        g <- gauge(f$JumpHeight_cm, min = min(t$JumpHeight_cm), max = max(t$JumpHeight_cm), symbol = 'cm', gaugeSectors(
            success = c((max(t$JumpHeight_cm)*.9), max(t$JumpHeight_cm)), warning = c((max(t$JumpHeight_cm)*.8), max(t$JumpHeight_cm)*.9), danger = c(min(t$JumpHeight_cm), max(t$JumpHeight_cm)*.8)
        ))
        print(g)
    })
    })

数据

jumpdata <- structure(list(Athlete = structure(c(1L, 1L, 1L, 7L, 7L, 7L, 
7L, 8L, 8L, 8L, 8L, 9L, 9L, 9L, 9L, 10L, 10L, 10L, 10L, 11L, 
11L, 11L, 11L, 12L, 12L, 12L, 12L, 13L, 13L, 13L, 13L, 14L, 14L, 
14L, 14L, 2L, 2L, 2L, 2L, 3L, 3L, 3L, 3L, 4L, 4L, 4L, 4L, 5L, 
5L, 5L, 5L, 6L, 6L, 6L, 6L), .Label = c("Athlete 1", "Athlete 10", 
"Athlete 11", "Athlete 12", "Athlete 13", "Athlete 14", "Athlete 2", 
"Athlete 3", "Athlete 4", "Athlete 5", "Athlete 6", "Athlete 7", 
"Athlete 8", "Athlete 9"), class = "factor"), Date = structure(c(1L, 
4L, 5L, 1L, 3L, 5L, 7L, 2L, 3L, 5L, 7L, 1L, 3L, 5L, 7L, 1L, 3L, 
5L, 7L, 1L, 3L, 6L, 7L, 2L, 4L, 5L, 8L, 1L, 3L, 5L, 7L, 1L, 3L, 
5L, 7L, 1L, 3L, 5L, 7L, 1L, 3L, 5L, 7L, 1L, 3L, 5L, 7L, 1L, 3L, 
6L, 7L, 1L, 3L, 5L, 7L), .Label = c("2020-01-06", "2020-01-07", 
"2020-01-13", "2020-01-14", "2020-01-21", "2020-01-23", "2020-01-27", 
"2020-01-28"), class = "factor"), Position = structure(c(2L, 
2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 1L, 1L, 1L, 1L, 2L, 2L, 
2L, 2L, 2L, 2L, 2L, 2L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 2L, 2L, 
2L, 2L, 2L, 2L, 2L, 2L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 3L, 3L, 
3L, 3L, 3L, 3L, 3L, 3L), .Label = c("DEF", "FWD", "GOALIE"), class = "factor"), 
    Program = structure(c(2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 
    2L, 2L, 1L, 1L, 1L, 1L, 4L, 4L, 4L, 4L, 4L, 4L, 4L, 4L, 1L, 
    1L, 1L, 1L, 2L, 2L, 2L, 2L, 4L, 4L, 4L, 4L, 1L, 1L, 1L, 1L, 
    4L, 4L, 4L, 4L, 1L, 1L, 1L, 1L, 3L, 3L, 3L, 3L, 3L, 3L, 3L, 
    3L), .Label = c("Navy", "Red", "RTP", "White"), class = "factor"), 
    mRSI = c(0.36, 0.38, 0.42, 0.46, 0.46, 0.47, 0.48, 0.31, 
    0.3, 0.24, 0.3, 0.29, 0.26, 0.28, 0.28, 0.36, 0.35, 0.43, 
    0.43, 0.28, 0.31, 0.28, 0.3, 0.33, 0.36, 0.35, 0.37, 0.37, 
    0.36, 0.37, 0.36, 0.3, 0.36, 0.34, 0.37, 0.26, 0.28, 0.34, 
    0.3, 0.39, 0.4, 0.43, 0.43, 0.43, 0.47, 0.46, 0.48, 0.34, 
    0.36, 0.33, 0.37, 0.28, 0.28, 0.34, 0.33), SystemWeight = c(617.21, 
    612.4, 620.45, 672.08, 682.23, 670.5, 663.41, 517.33, 515.23, 
    511.62, 517.85, 697.55, 703.92, 689.43, 691.33, 859.06, 845.9, 
    850.97, 851.84, 655.79, 665.09, 673.91, 667.92, 626.78, 632.92, 
    634.52, 624.88, 637.55, 645.6, 648.78, 646.64, 558.03, 563.23, 
    569.58, 560.95, 693.63, 695.54, 684.37, 684.58, 641.18, 660.8, 
    663.95, 660, 594.92, 596.97, 591.36, 585.64, 522.35, 518.17, 
    530.95, 523.5, 780.65, 789.81, 775.84, 775.48), FTCT = c(0.61, 
    0.62, 0.67, 0.74, 0.75, 0.77, 0.77, 0.54, 0.55, 0.44, 0.53, 
    0.53, 0.49, 0.53, 0.56, 0.6, 0.58, 0.68, 0.68, 0.53, 0.57, 
    0.54, 0.55, 0.61, 0.63, 0.64, 0.65, 0.59, 0.58, 0.59, 0.59, 
    0.51, 0.59, 0.59, 0.59, 0.53, 0.57, 0.63, 0.59, 0.76, 0.76, 
    0.79, 0.78, 0.67, 0.72, 0.72, 0.74, 0.63, 0.65, 0.61, 0.63, 
    0.49, 0.5, 0.53, 0.57), JumpHeight_cm = c(28.97, 29.78, 31.43, 
    35.83, 35.41, 36.59, 36.92, 27.56, 26.11, 26.15, 26.82, 26.15, 
    25.08, 24.98, 24.62, 29.39, 30.17, 32.42, 32.56, 26.6, 27.25, 
    25.58, 27.88, 29.17, 31.58, 28.48, 31.24, 33.73, 32.78, 33.09, 
    33.43, 29.73, 31.91, 30.65, 32.98, 24.15, 24.24, 27.57, 25.44, 
    26.68, 26.39, 27.43, 28.87, 35.44, 36.29, 35.71, 36.06, 26.79, 
    27.76, 26.82, 29.71, 28.69, 26.9, 31.12, 29.77), EJH = c(17.6, 
    18.58, 21.11, 26.66, 26.69, 28.08, 28.38, 14.99, 14.39, 11.41, 
    14.33, 13.8, 12.34, 13.29, 13.67, 17.58, 17.5, 22.03, 22.19, 
    14.03, 15.59, 13.92, 15.39, 17.7, 19.75, 18.37, 20.3, 19.99, 
    18.9, 19.62, 19.61, 15.09, 18.8, 18.18, 19.6, 12.78, 13.87, 
    17.28, 15.06, 20.44, 20.12, 21.74, 22.52, 23.8, 26.25, 25.68, 
    26.73, 16.99, 18.13, 16.42, 18.82, 14.09, 13.43, 16.61, 16.9
    ), Weight = c(62.94, 62.45, 63.27, 68.54, 69.57, 68.38, 67.65, 
    52.76, 52.54, 52.17, 52.81, 71.13, 71.78, 70.31, 70.5, 87.61, 
    86.26, 86.78, 86.87, 66.88, 67.82, 68.72, 68.11, 63.92, 64.54, 
    64.71, 63.72, 65.02, 65.84, 66.16, 65.94, 56.91, 57.44, 58.09, 
    57.2, 70.74, 70.93, 69.79, 69.81, 65.39, 67.39, 67.71, 67.31, 
    60.67, 60.88, 60.31, 59.72, 53.27, 52.84, 54.15, 53.39, 79.61, 
    80.54, 79.12, 79.08)), class = "data.frame", row.names = c(NA, 
-55L))

根据发布在github上的解决方法,这是我的新代码,但无法渲染。我不确定基于我的原始量表要包含什么作为input$range

ui.r

library(shiny)
library(shinydashboard)
library(flexdashboard)
library(dplyr)

jumpdata <- read.csv("SO CMJ Dummy.csv")
jumpdata$Date <- as.Date(jumpdata$Date, "%Y-%m-%d")

shinyUI(
    fluidPage(
        sidebarPanel(width = 3,
                     selectInput("Athlete", label = "Athlete",
                                 choices = unique(jumpdata$Athlete))),
        mainPanel(
            fluidRow(
                box(title = "Jump Height", gaugeOutput("Gauge_JH")),
                uiOutput("Gauge_JH_Proxy")
            ))
    ))

server.r

library(shiny)
library(shinydashboard)
library(flexdashboard)
library(dplyr)

jumpdata <- read.csv("SO CMJ Dummy.csv")
jumpdata$Date <- as.Date(jumpdata$Date, "%Y-%m-%d")

shinyServer(function(input, output){
    output$Gauge_JH <- renderGauge({
        f <- jumpdata %>%
            select(Date, Athlete, JumpHeight_cm) %>%
            filter(Athlete == input$Athlete & Date == c(max(jumpdata$Date)))

        t <- jumpdata %>%
            select(Date, Athlete, JumpHeight_cm) %>%
            filter(Athlete == input$Athlete)

        g <- gauge(f$JumpHeight_cm, min = min(t$JumpHeight_cm), max = max(t$JumpHeight_cm), symbol = 'cm', gaugeSectors(
            success = c((max(t$JumpHeight_cm)*.9), max(t$JumpHeight_cm)), warning = c((max(t$JumpHeight_cm)*.8), max(t$JumpHeight_cm)*.9), danger = c(min(t$JumpHeight_cm), max(t$JumpHeight_cm)*.8)
        ))
        print(g)
    })
    output$Gauge_JH_Proxy <- renderUI({

        input$Athlete # force re-rendering
        gaugeOutput(outputId = "Gauge_JH", width = "30%", height = "200px")
    }) 

})
r shiny reactive flexdashboard gauge
1个回答
0
投票

可以通过使用renderUIdebounce解决此问题(延迟渲染,以使计算准备就绪。)>

请注意,我已更改范围逻辑以实际显示一些颜色并查看以下内容:

library(shiny)
library(shinydashboard)
library(flexdashboard)
library(dplyr)

jumpdata <- structure(list(Athlete = structure(c(1L, 1L, 1L, 7L, 7L, 7L, 
                                                 7L, 8L, 8L, 8L, 8L, 9L, 9L, 9L, 9L, 10L, 10L, 10L, 10L, 11L, 
                                                 11L, 11L, 11L, 12L, 12L, 12L, 12L, 13L, 13L, 13L, 13L, 14L, 14L, 
                                                 14L, 14L, 2L, 2L, 2L, 2L, 3L, 3L, 3L, 3L, 4L, 4L, 4L, 4L, 5L, 
                                                 5L, 5L, 5L, 6L, 6L, 6L, 6L), .Label = c("Athlete 1", "Athlete 10", 
                                                                                         "Athlete 11", "Athlete 12", "Athlete 13", "Athlete 14", "Athlete 2", 
                                                                                         "Athlete 3", "Athlete 4", "Athlete 5", "Athlete 6", "Athlete 7", 
                                                                                         "Athlete 8", "Athlete 9"), class = "factor"), Date = structure(c(1L, 
                                                                                                                                                          4L, 5L, 1L, 3L, 5L, 7L, 2L, 3L, 5L, 7L, 1L, 3L, 5L, 7L, 1L, 3L, 
                                                                                                                                                          5L, 7L, 1L, 3L, 6L, 7L, 2L, 4L, 5L, 8L, 1L, 3L, 5L, 7L, 1L, 3L, 
                                                                                                                                                          5L, 7L, 1L, 3L, 5L, 7L, 1L, 3L, 5L, 7L, 1L, 3L, 5L, 7L, 1L, 3L, 
                                                                                                                                                          6L, 7L, 1L, 3L, 5L, 7L), .Label = c("2020-01-06", "2020-01-07", 
                                                                                                                                                                                              "2020-01-13", "2020-01-14", "2020-01-21", "2020-01-23", "2020-01-27", 
                                                                                                                                                                                              "2020-01-28"), class = "factor"), Position = structure(c(2L, 
                                                                                                                                                                                                                                                       2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 1L, 1L, 1L, 1L, 2L, 2L, 
                                                                                                                                                                                                                                                       2L, 2L, 2L, 2L, 2L, 2L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 2L, 2L, 
                                                                                                                                                                                                                                                       2L, 2L, 2L, 2L, 2L, 2L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 3L, 3L, 
                                                                                                                                                                                                                                                       3L, 3L, 3L, 3L, 3L, 3L), .Label = c("DEF", "FWD", "GOALIE"), class = "factor"), 
                           Program = structure(c(2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 
                                                 2L, 2L, 1L, 1L, 1L, 1L, 4L, 4L, 4L, 4L, 4L, 4L, 4L, 4L, 1L, 
                                                 1L, 1L, 1L, 2L, 2L, 2L, 2L, 4L, 4L, 4L, 4L, 1L, 1L, 1L, 1L, 
                                                 4L, 4L, 4L, 4L, 1L, 1L, 1L, 1L, 3L, 3L, 3L, 3L, 3L, 3L, 3L, 
                                                 3L), .Label = c("Navy", "Red", "RTP", "White"), class = "factor"), 
                           mRSI = c(0.36, 0.38, 0.42, 0.46, 0.46, 0.47, 0.48, 0.31, 
                                    0.3, 0.24, 0.3, 0.29, 0.26, 0.28, 0.28, 0.36, 0.35, 0.43, 
                                    0.43, 0.28, 0.31, 0.28, 0.3, 0.33, 0.36, 0.35, 0.37, 0.37, 
                                    0.36, 0.37, 0.36, 0.3, 0.36, 0.34, 0.37, 0.26, 0.28, 0.34, 
                                    0.3, 0.39, 0.4, 0.43, 0.43, 0.43, 0.47, 0.46, 0.48, 0.34, 
                                    0.36, 0.33, 0.37, 0.28, 0.28, 0.34, 0.33), SystemWeight = c(617.21, 
                                                                                                612.4, 620.45, 672.08, 682.23, 670.5, 663.41, 517.33, 515.23, 
                                                                                                511.62, 517.85, 697.55, 703.92, 689.43, 691.33, 859.06, 845.9, 
                                                                                                850.97, 851.84, 655.79, 665.09, 673.91, 667.92, 626.78, 632.92, 
                                                                                                634.52, 624.88, 637.55, 645.6, 648.78, 646.64, 558.03, 563.23, 
                                                                                                569.58, 560.95, 693.63, 695.54, 684.37, 684.58, 641.18, 660.8, 
                                                                                                663.95, 660, 594.92, 596.97, 591.36, 585.64, 522.35, 518.17, 
                                                                                                530.95, 523.5, 780.65, 789.81, 775.84, 775.48), FTCT = c(0.61, 
                                                                                                                                                         0.62, 0.67, 0.74, 0.75, 0.77, 0.77, 0.54, 0.55, 0.44, 0.53, 
                                                                                                                                                         0.53, 0.49, 0.53, 0.56, 0.6, 0.58, 0.68, 0.68, 0.53, 0.57, 
                                                                                                                                                         0.54, 0.55, 0.61, 0.63, 0.64, 0.65, 0.59, 0.58, 0.59, 0.59, 
                                                                                                                                                         0.51, 0.59, 0.59, 0.59, 0.53, 0.57, 0.63, 0.59, 0.76, 0.76, 
                                                                                                                                                         0.79, 0.78, 0.67, 0.72, 0.72, 0.74, 0.63, 0.65, 0.61, 0.63, 
                                                                                                                                                         0.49, 0.5, 0.53, 0.57), JumpHeight_cm = c(28.97, 29.78, 31.43, 
                                                                                                                                                                                                   35.83, 35.41, 36.59, 36.92, 27.56, 26.11, 26.15, 26.82, 26.15, 
                                                                                                                                                                                                   25.08, 24.98, 24.62, 29.39, 30.17, 32.42, 32.56, 26.6, 27.25, 
                                                                                                                                                                                                   25.58, 27.88, 29.17, 31.58, 28.48, 31.24, 33.73, 32.78, 33.09, 
                                                                                                                                                                                                   33.43, 29.73, 31.91, 30.65, 32.98, 24.15, 24.24, 27.57, 25.44, 
                                                                                                                                                                                                   26.68, 26.39, 27.43, 28.87, 35.44, 36.29, 35.71, 36.06, 26.79, 
                                                                                                                                                                                                   27.76, 26.82, 29.71, 28.69, 26.9, 31.12, 29.77), EJH = c(17.6, 
                                                                                                                                                                                                                                                            18.58, 21.11, 26.66, 26.69, 28.08, 28.38, 14.99, 14.39, 11.41, 
                                                                                                                                                                                                                                                            14.33, 13.8, 12.34, 13.29, 13.67, 17.58, 17.5, 22.03, 22.19, 
                                                                                                                                                                                                                                                            14.03, 15.59, 13.92, 15.39, 17.7, 19.75, 18.37, 20.3, 19.99, 
                                                                                                                                                                                                                                                            18.9, 19.62, 19.61, 15.09, 18.8, 18.18, 19.6, 12.78, 13.87, 
                                                                                                                                                                                                                                                            17.28, 15.06, 20.44, 20.12, 21.74, 22.52, 23.8, 26.25, 25.68, 
                                                                                                                                                                                                                                                            26.73, 16.99, 18.13, 16.42, 18.82, 14.09, 13.43, 16.61, 16.9
                                                                                                                                                                                                   ), Weight = c(62.94, 62.45, 63.27, 68.54, 69.57, 68.38, 67.65, 
                                                                                                                                                                                                                 52.76, 52.54, 52.17, 52.81, 71.13, 71.78, 70.31, 70.5, 87.61, 
                                                                                                                                                                                                                 86.26, 86.78, 86.87, 66.88, 67.82, 68.72, 68.11, 63.92, 64.54, 
                                                                                                                                                                                                                 64.71, 63.72, 65.02, 65.84, 66.16, 65.94, 56.91, 57.44, 58.09, 
                                                                                                                                                                                                                 57.2, 70.74, 70.93, 69.79, 69.81, 65.39, 67.39, 67.71, 67.31, 
                                                                                                                                                                                                                 60.67, 60.88, 60.31, 59.72, 53.27, 52.84, 54.15, 53.39, 79.61, 
                                                                                                                                                                                                                 80.54, 79.12, 79.08)), class = "data.frame", row.names = c(NA, 
                                                                                                                                                                                                                                                                            -55L))
jumpdata$Date <- as.Date(jumpdata$Date, "%Y-%m-%d")

ui <- fluidPage(
  fluidPage(
    sidebarPanel(width = 3,
                 selectInput("Athlete", label = "Athlete",
                             choices = unique(jumpdata$Athlete))),
    mainPanel(
      fluidRow(
        box(title = "Jump Height", uiOutput("Gauge_JH_Proxy"))
      ))
))

server <- function(input, output, session) {
    output$Gauge_JH <- renderGauge({
      g()
    })

    Athlete <- debounce(reactive({input$Athlete}), 500)

    output$Gauge_JH_Proxy <- renderUI({
      req(Athlete()) # force rerendering
      gaugeOutput("Gauge_JH") 
      })

    g <- reactive({
      t <- jumpdata %>%
        select(Date, Athlete, JumpHeight_cm) %>%
        filter(Athlete == input$Athlete)

      f <- t %>% filter(Date == max(Date))

      minJump = min(t$JumpHeight_cm)
      maxJump = max(t$JumpHeight_cm)
      diffJump = maxJump-minJump

        gauge(
          value = f$JumpHeight_cm,
          min = min(t$JumpHeight_cm),
          max = max(t$JumpHeight_cm),
          sectors = gaugeSectors(
            success = c(min(t$JumpHeight_cm) + diffJump * 0.8, max(t$JumpHeight_cm)),
            warning = c(min(t$JumpHeight_cm) + diffJump * 0.4, min(t$JumpHeight_cm) + diffJump * 0.8),
            danger = c(min(t$JumpHeight_cm), min(t$JumpHeight_cm) + diffJump * 0.4)
          ),
          symbol = 'cm'
        )
    })

}

shinyApp(ui, server)

Result shinydashboard

但是,在所有这些不便之处,我都会切换图书馆。这是一种plotly方法:

library(shiny)
library(shinydashboard)
library(dplyr)
library(plotly)

# jumpdata <- [copy & paste jumpdata here]
jumpdata$Date <- as.Date(jumpdata$Date, "%Y-%m-%d")

ui <- fluidPage(
  fluidPage(
    sidebarPanel(width = 3,
                 selectInput("Athlete", label = "Athlete",
                             choices = unique(jumpdata$Athlete))),
    mainPanel(
      fluidRow(
        plotlyOutput("Gauge_JH_plotly", height = 250, width = "50%")
      ))
))

server <- function(input, output, session) {

    output$Gauge_JH_plotly <- renderPlotly({

      t <- jumpdata %>%
        select(Date, Athlete, JumpHeight_cm) %>%
        filter(Athlete == input$Athlete)

      f <- t %>% filter(Date == max(Date))

      currentJump = f$JumpHeight_cm
      meanJump = mean(t$JumpHeight_cm)
      minJump = min(t$JumpHeight_cm)
      maxJump = max(t$JumpHeight_cm)
      diffJump = maxJump-minJump

      success = c(min(t$JumpHeight_cm) + diffJump * 0.8, max(t$JumpHeight_cm))
      warning = c(min(t$JumpHeight_cm) + diffJump * 0.4, min(t$JumpHeight_cm) + diffJump * 0.8)
      danger = c(min(t$JumpHeight_cm), min(t$JumpHeight_cm) + diffJump * 0.4)

      ranges <- unique(c(danger, warning, success))

      currentJumpColor <- c("red", "orange", "green")[findInterval(currentJump, ranges, rightmost.closed = TRUE)]

      fig <- plot_ly(
        domain = list(x = c(0, 1), y = c(0, 1)),
        value = currentJump,
        title = list(text = "Jump Height [cm]"),
        type = "indicator",
        mode = "gauge+number+delta",
        delta = list(reference = meanJump),
        gauge = list(
          bar = list(color = currentJumpColor),
          axis = list(range = list(minJump, maxJump)),
          steps = list(
            list(range = danger, color = "lightgray"),
            list(range = warning, color = "gray")),
          threshold = list(
            line = list(color = "green", width = 4),
            thickness = 0.75,
            value = maxJump))) 
      fig <- fig %>% layout(margin = list(l=30, r=30, t=80, b=30))

      fig
    })

}


shinyApp(ui, server)

Result plotly

© www.soinside.com 2019 - 2024. All rights reserved.