[R] R gets slow
pau carre
pau.carre at gmail.com
Sat Mar 25 21:16:32 CET 2006
Hello, I have R as a socket server that computes R code sent by some
scripts (the clients). These scrips send R code to generate models
(SVM). The problem is that first models are generated in less than one
second and one hour later, the same models are generated in more than
ten seconds (even training with same data). If I restart the server ,
then it works well (fast). I don't know if I have to free the memory
or something.
Here you have the code:
R server:
FSsocket <- function(){
continue = TRUE;
while(continue){
conn <- try(socketConnection(server = TRUE, port = 7890, blocking =
TRUE, open = "ab"), silent = FALSE);
isOpened = !inherits(conn, "try-error");
isOpened = isOpened && isOpen(conn);
while(!isOpened){
Sys.sleep(1);
conn <- try(socketConnection(server = TRUE, port = 7890, blocking
= TRUE, open = "ab"), silent = FALSE);
isOpened = !inherits(conn, "try-error");
isOpened = isOpened && isOpen(conn);
}
print("Waitting for source");
srcFile = readLines(conn, n = 1)
print(srcFile)
continue = srcFile != "--close"
if(continue){
print("Executing source");
error <- try(source(srcFile), silent = FALSE);
if(inherits(error, "try-error")){
writeLines("ERROR", conn)
}
print("Sending confirmation")
writeLines("DONE", conn)
print("Closing connection")
}
close(conn)
}
}
FSsocket();
Model generator example:
library("class");
library("e1071");
dd = read.table("sampling/adapteddataSet");
attach(dd);
ddv = read.table("sampling/adaptedvalidationDataSet");
attach(ddv);
dd[,1] = factor(dd[,1]);
ddv[,1] = factor(ddv[,1]);
attach(dd);
tr_in = as.matrix(dd[,2:(1 + 1)]);
tr_out = dd[,1];
val_in = as.matrix(ddv[,2:(1 + 1)]);
val_out = ddv[,1];
t = tune(svm, kernel = "radial", train.x = tr_in, train.y = tr_out,
validation.x = val_in, validation.y = val_out,
range = list( gamma = 2^(-1:1), cost = 2^(2:4) ), tunecontrol =
tune.control(sampling = "fix") )
z = t$best.model
save(z, file = "./models/1/20", compress=FALSE);
Thanks
Pau.
More information about the R-help
mailing list