使用并行计算在 R 中保存地块

Question

rm(list=ls())
    analysis <- function(time_series,sublist){
    ROIs <- c("RW","RFHM","RFD2","RFD3","RFD4")
    data <- time_series[[sublist]]
    for (subject in 1:length(data$successful_x_direction)){
        subject_name <- paste0("subject_",subject)
        success_x <- data$successful_x_direction[[subject]]
        success_y <- data$successful_y_direction[[subject]]
        fail_x <- data$failure_x_direction[[subject]]
        fail_y <- data$failure_y_direction[[subject]]
        for (success in 1:length(success_x)){
            matrix_x <- success_x[[success]]
            matrix_y <- success_y[[success]]
            matrix_x <- as.data.frame(matrix_x)
            matrix_x$Frame <- seq(1,dim(matrix_x)[1])
            matrix_x <- matrix_x %>% pivot_longer(!Frame,names_to = "ROI",values_to = "Value")
            x_cor <- ggplot(matrix_x,aes(x = Frame, y = Value, colour = ROI)) + geom_line() + ylab("X Coordinates")
            matrix_y <- as.data.frame(matrix_y)
            matrix_y$Frame <- seq(1,dim(matrix_y)[1])
            matrix_y <- matrix_y %>% pivot_longer(!Frame,names_to = "ROI",values_to = "Value")
            y_cor <- ggplot(matrix_y,aes(x = Frame, y = Value, colour = ROI)) + geom_line() +  ylab("Y Coordinates")
            g <- ggarrange(x_cor,y_cor,ncol = 1, nrow = 2)
            g <- annotate_figure(g, top = text_grob("Success", color = "red", face = "bold", size = 14))
           filename = paste0("Success_",success,".png")
           filename = paste("Time Series Plots",sublist,subject_name,filename,sep = "\\")
           ggsave(plot = g, filename = filename)
    }
}

Hello, I have this code which basically saves some plots. Using a traditional for loop works fine:

load("time_series.Rdata")
for (i in seq(1,length(time_series))){
    sublist <- names(time_series)[i]
    analysis(time_series,sublist)
}

but when I try to run it in parallel it doesn't work. I wait for several minutes and nothing is being saved and my CPU is not being used. What am I doing wrong?

library(doParallel)
library(foreach)
numCores <- detectCores()-1
cl <- makeCluster(numCores)
registerDoParallel(cl)
random_name <- foreach(i=seq(1,length(time_series))) %dopar% {
    sublist <- names(time_series)[i]
    analysis(time_series,sublist)
}
stopCluster(cl)

Answer 1

you should provide a reprex if there is something peculiar about your data; without a reprex I can only try to make something simple with similar structure to show in principle if it can work. I use tictoc for timing; on my machine normal runtime is 2seconds; parallel is slower 2.7seconds.

library(tidyverse)
library(tictoc)

data_to_use <- list(a1=iris,
                    b2=mutate(iris,across(where(is.numeric),~.x*.x)))

analysis <- function(df,nm){
   require(ggplot2)
   require(dplyr)
  for(i in unique(df$Species)){
    g <- ggplot(data=df |> filter(Species==i),
                aes(x=Petal.Length,
                    y=Petal.Width))+geom_point()
    ggsave(plot = g, 
           filename = paste0(nm,"_",i,".png"))
  }
}

tic()
analysis(df=data_to_use[[1]],names(data_to_use)[[1]])
analysis(df=data_to_use[[2]],names(data_to_use)[[2]])
toc()


library(doParallel)
library(foreach)
numCores <- detectCores()-1
cl <- makeCluster(numCores)
registerDoParallel(cl)

tic()
foreach(i=1:2,.packages = "tictoc",) %dopar% {

  analysis(data_to_use[[i]],
           names(data_to_use)[[i]])

}
toc()
stopCluster(cl)

Answer 2

It seems that my script was working fine. I just had to wait more than expected for the plots to be saved.