Error during installation R GPU win64: 'R' is not recognized as an internal or external command, operable program or batch file

marboe123 · October 5, 2021, 1:38pm

I try to install R GPU win64 from:

https://s3-us-west-2.amazonaws.com/xgboost-nightly-builds/release_1.5.0/xgboost_r_gpu_win64_508a0b0dbd674909e8ec53881e17f3363fd9b508.tar.gz

Although if I enter: R CMD INSTALL ./xgboost_r_gpu_linux.tar.gz

I receive this error:
‘R’ is not recognized as an internal or external command,
operable program or batch file.

This is my setup:
R version 4.1.0 (2021-05-18) – “Camp Pontanezen”
Rtools installed.
Windows 10 64 bit.

I see these solutions or the error:

Although my line above does not have spaces so I suppose the root cause is different in my case.

Any idea what can be the cause of the error?

Thank you!

hcho3 · October 5, 2021, 10:10pm

Please check your PATH environment variable. https://support.shotgunsoftware.com/hc/en-us/articles/114094235653-Setting-global-environment-variables-on-Windows

marboe123 · October 6, 2021, 9:30am

These are my current variables:

I see that the Path variabele is related to NVIDIA GPU Computing Toolkit. If I remember correctly I have set this during installation of Keras for deep learning in R.

Do you know if I have to replace this value?

Thank you!

hcho3 · October 6, 2021, 4:21pm

@marboe123 You should add the folder that contains the executable R. For example, if the R executable is found in C:\Program Files\R\R-4.1.0\bin, then the PATH should be revised to include C:\Program Files\R\R-4.1.0\bin. Do not delete other folders that were already in PATH; it is sufficient to add the R folder to PATH.

marboe123 · October 7, 2021, 10:16am

@hcho3

Thank you very much!

I did succeed to install, although during installation there are a few lines which has a [ FAIL ] at the end of the line. I have copied the output I received during installation below.

I also receive an error if I try to run xgb.train. See my output at the end of this post.

C:\xgboost_gpu_file>R CMD INSTALL ./xgboost_r_gpu_win64_508a0b0dbd674909e8ec53881e17f3363fd9b508.tar.gz

installing to library ‘C:/Users/boers/Documents/R/win-library/4.1’
installing source package ‘xgboost’ …
** using staged installation
** libs
running ‘src/Makefile.win’ …
make: Nothing to be done for ‘all’.
installing to C:/Users/boers/Documents/R/win-library/4.1/00LOCK-xgboost/00new/xgboost/libs/x64
** R
** data
** demo
** inst
** byte-compile and prepare package for lazy loading
** help
*** installing help indices
converting help for package ‘xgboost’
finding HTML links … done
a-compatibility-note-for-saveRDS-save html
agaricus.test html
agaricus.train html
callbacks html
cb.cv.predict html
cb.early.stop html
cb.evaluation.log html
cb.gblinear.history html
cb.print.evaluation html
cb.reset.parameters html
cb.save.model html
dim.xgb.DMatrix html
dimnames.xgb.DMatrix html
REDIRECT:topic dimnames<-.xgb.DMatrix -> dimnames.xgb.DMatrix.html [ FAIL ]
getinfo html
normalize html
predict.xgb.Booster html
prepare.ggplot.shap.data html
print.xgb.Booster html
print.xgb.DMatrix html
print.xgb.cv html
setinfo html
slice.xgb.DMatrix html
xgb.Booster.complete html
xgb.DMatrix html
xgb.DMatrix.save html
xgb.attr html
REDIRECT:topic xgb.attr<- -> xgb.attr.html [ FAIL ]
REDIRECT:topic xgb.attributes<- -> xgb.attr.html [ FAIL ]
xgb.config html
REDIRECT:topic xgb.config<- -> xgb.config.html [ FAIL ]
xgb.create.features html
xgb.cv html
xgb.dump html
xgb.gblinear.history html
xgb.importance html
xgb.load html
xgb.load.raw html
xgb.model.dt.tree html
xgb.parameters html
REDIRECT:topic xgb.parameters<- -> xgb.parameters.html [ FAIL ]
xgb.plot.deepness html
xgb.plot.importance html
xgb.plot.multi.trees html
xgb.plot.shap html
xgb.plot.shap.summary html
xgb.plot.tree html
xgb.save html
xgb.save.raw html
xgb.serialize html
xgb.shap.data html
xgb.train html
xgb.unserialize html
xgbConfig html
xgboost-deprecated html
** building package indices
** installing vignettes
** testing if installed package can be loaded from temporary location
** testing if installed package can be loaded from final location
** testing if installed package keeps a record of temporary installation path
DONE (xgboost)

Next I have added “tree_method = gpu_hist” to xgb.train as in my code below, although this gives the error:

Error in xgb.train(data = xgb_trainval, tree_method = gpu_hist, booster = “gbtree”, :
object ‘gpu_hist’ not found

  model_n <- xgb.train(data = xgb_trainval,
                     tree_method = gpu_hist,
                     booster = "gbtree",
                     objective = "binary:logistic",
                     max_depth = parameters_df$max_depth[row],
                     eta = parameters_df$eta[row],
                     subsample = parameters_df$subsample[row],
                     colsample_bytree = parameters_df$colsample_bytree[row],
                     min_child_weight = parameters_df$min_child_weight[row],
                     nrounds = 300,
                     eval_metric = "auc",
                     early_stopping_rounds = 30,
                     print_every_n = 100,
                     watchlist = list(train = xgb_trainval, val = xgb_val)

marboe123 · October 7, 2021, 11:00am

I did find the cause of the error : object ‘gpu_hist’ not found
I had forgot to use quotes in:

tree_method = “gpu_hist”

Great, it is running now!

marboe123 · October 11, 2021, 12:37pm

I have compared the outcomes of the same code (which calculates AUC for several hyperparametersettings) on CPU and GPU with the same seed just before

xgb.train

I notice that the outcomes show a high correlation although they are not exactly identical.

I suppose this is acceptable and due to some randomness in the proces. Although I just want to report it in case it should be exactly identical.

The processing time shows an improvement from 167.4 minutes to 26.0 minutes (6.44x) which is great imo.

hcho3 · October 12, 2021, 6:30am

No, there is no expectation that CPU and GPU algorithms would produce exactly identical results.