Mor*_*erg 5 parallel-processing r shiny
我正在为我创建的模拟器创建一个闪亮的应用程序.为了加快模拟速度,我使用了parallel包.
我的应用程序在没有并行化我的代码时工作正常,尽管它很慢.但是,当我并行化时,我收到以下错误:
Error in checkForRemoteErrors(val) :
3 nodes produced errors; first error: Operation not allowed without an active reactive context. (You tried to do something that can only be done from inside a reactive expression or observer.)
这是我的ui.R和server.R的删节版本:
library(shiny)
shinyUI(fluidPage(
titlePanel("Simulator"),
fluidRow(
column(6,
fluidRow(
column(5,
helpText("Choose 9 bitcoins for firm 1"),
selectizeInput("firm1bit1", label = "Bitcoin 1:",
choices = data$bitcoin, options =
list(maxOptions = 7)),
selectizeInput("firm1bit2", label = "Bitcoin 2:",
choices = data$bitcoin, options =
list(maxOptions = 7)),
selectizeInput("firm1bit3", label = "Bitcoin 3:",
choices = data$bitcoin, options =
list(maxOptions = 7)),
selectizeInput("firm1bit4", label = "Bitcoin 4:",
choices = data$bitcoin, options =
list(maxOptions = 7)),
selectizeInput("firm1bit5", label = "Bitcoin 5:",
choices = data$bitcoin, options =
list(maxOptions = 7)),
selectizeInput("firm1bit6", label = "Bitcoin 6:",
choices = data$bitcoin, options =
list(maxOptions = 7)),
selectizeInput("firm1bit7", label = "Bitcoin 7:",
choices = data$bitcoin, options =
list(maxOptions = 7)),
selectizeInput("firm1bit8", label = "Bitcoin 8:",
choices = data$bitcoin, options =
list(maxOptions = 7)),
selectizeInput("firm1bit9", label = "Bitcoin 9:",
choices = data$bitcoin, options =
list(maxOptions = 7)),
helpText("Choose the maximum number of transactions for firm 1"),
selectizeInput("firm1transacts", label = "Firm 1 maximum number of transactions:",
choices = data$max_transactions, options =
list(maxOptions = 7))
),
column(5,
helpText("Choose 9 bitcoins for firm 2"),
selectizeInput("firm2bit1", label = "Bitcoin 1:",
choices = data$bitcoin, options =
list(maxOptions = 7)),
selectizeInput("firm2bit2", label = "Bitcoin 2:",
choices = data$bitcoin, options =
list(maxOptions = 7)),
selectizeInput("firm2bit3", label = "Bitcoin 3:",
choices = data$bitcoin, options =
list(maxOptions = 7)),
selectizeInput("firm2bit4", label = "Bitcoin 4:",
choices = data$bitcoin, options =
list(maxOptions = 7)),
selectizeInput("firm2bit5", label = "Bitcoin 5:",
choices = data$bitcoin, options =
list(maxOptions = 7)),
selectizeInput("firm2bit6", label = "Bitcoin 6:",
choices = data$bitcoin, options =
list(maxOptions = 7)),
selectizeInput("firm2bit7", label = "Bitcoin 7:",
choices = data$bitcoin, options =
list(maxOptions = 7)),
selectizeInput("firm2bit8", label = "Bitcoin 8:",
choices = data$bitcoin, options =
list(maxOptions = 7)),
selectizeInput("firm2bit9", label = "Bitcoin 9:",
choices = data$bitcoin, options =
list(maxOptions = 7)),
helpText("Choose the maximum number of transactions for firm 2"),
selectizeInput("firm2transacts", label = "Firm 2 maximum number of transactions:",
choices = data$max_transactions, options =
list(maxOptions = 7))
),
submitButton("Simulate")
))
)
))
cl <- makeCluster(detectCores()-1, 'PSOCK')
shinyServer(function(input, output, session){
firm1bits <- reactive({c(input$firm1bit1, input$firm1bit2, input$firm1bit3,
input$firm1bit4, input$firm1bit5, input$firm1bit6,
input$firm1bit7, input$firm1bit8, input$firm1bit9)})
firm2bits <- reactive({c(input$firm2bit1, input$firm2bit2, input$firm2bit3,
input$firm2bit4, input$firm2bit5, input$firm2bit6,
input$firm2bit7, input$firm2bit8, input$firm2bit9)})
firm1max <- reactive({input$firm1transacts})
firm2max <- reactive({input$firm2transacts})
reactive({clusterExport(cl, varlist=c("firm1bits", "firm2bits", "firm1max",
"firm2max"))})
gameResults <- reactive({parSapply(cl, 1:1000, function(i){
simulate_bitcoin_Twoway(firm1bits(), firm2bits(), firm1max(), firm2max())
})})
})
我想重申代码在我不使用时使用parSapply()而是使用replicate().问题不在于其他功能,例如simulate_bitcoin_Twoway().
由于你没有提供MCVE,因此它比其他任何东西都更加狂野.
当您调用时clusterExport,在集群上分配响应变量.在集群上parSapply执行simulate_bitcoin_Twoway,为每个工作者提供单独的环境,而不包含reactive块.由于无功值需要反应上下文,因此整个操作失败.
为了解决这个问题,我会尝试在本地评估反应式表达式并分配返回值:
gameResults <- reactive({
firm1bits_v <- firm1bits()
firm2bits_v <- firm2bits()
firm1max_v <- firm1max()
firm2max_v <- firm2max()
clusterExport(cl, varlist=c(
"firm1bits_v", "firm2bits_v", "firm1max_v", "firm2max_v"))
parSapply(cl, 1:1000, function(i ){
simulate_bitcoin_Twoway(firm1bits_v, firm2bits_v, firm1max_v, firm2max_v)
})
})
如果上面不起作用,您可以尝试依赖于反应值,但在isolate块内部的集群上进行评估.
编辑:
这是一个完整的工作示例:
library(shiny)
library(parallel)
library(ggplot2)
cl <- makeCluster(detectCores()-1, 'PSOCK')
sim <- function(x, y, z) {
c(rnorm(1, mean=x), rnorm(1, mean=y), rnorm(1, mean=z))
}
shinyApp(
ui=shinyUI(bootstrapPage(
numericInput("x", "x", 10, min = 1, max = 100),
numericInput("y", "y", 10, min = 1, max = 100),
numericInput("z", "z", 10, min = 1, max = 100),
plotOutput("plot")
)),
server=shinyServer(function(input, output, session){
output$plot <- renderPlot({
x <- input$x
y <- input$y
z <- input$z
clusterExport(
cl, varlist=c("x", "y", "z", "sim"),
envir=environment())
mat <- t(parSapply(cl, 1:1000, function(i) {
sim(x, y, z)
}))
ggplot(
as.data.frame(mat),
aes(x=V1, y=V2, col=cut(V3, breaks=10))) + geom_point()
})
})
)
请注意envir参数clusterExport.默认情况下clusterExport,在全局环境中搜索,其中闭包中定义的变量不可见.