Dear all,
I have a script that worked well in RStudio on windows and RStudio server on a linux HPC. It tunes a multi-layer perceptron with ParBayesianOptimization. I tried to make it work in parallel, this way:
library(doParallel)
cl <- makeCluster(2)
registerDoParallel(cl)
clusterExport(cl,c('OUT','X_trII', 'y_trII',"Xtr_val_data","y_trn_val_data",'my_pearson_corr','my_negative_pearson_corr','accuracy_metric'))
clusterEvalQ(cl,expr= { library(keras) ; library(reticulate)})
optObj <- bayesOpt(
FUN = MLP_model_to_tune,
bounds = bounds,
initPoints = 7,
iters.n = 4,
iters.k = 2,
kappa=2.56,eps=0,
parallel=TRUE,
errorHandling = "continue",
plotProgress = FALSE, verbose = 1)
stopCluster(cl)
registerDoSEQ()
but it fails:
Running initial scoring function 7 times in 2 thread(s)... 11.89 seconds
dropout1 dropout2 units_Fl percent_dec nDenseLayers learning_rate
1: 0.34923411 0.29540303 278.37521 0.20122780 2.442593 0.073281385
2: 0.48399678 0.54744284 265.57111 0.48144830 13.217194 0.024877648
3: 0.06770692 0.07871296 183.41782 0.01590757 11.387769 0.001059684
4: 0.53342713 0.18634436 195.52110 0.21755177 15.370336 0.061193132
5: 0.17493806 0.44141748 36.67472 0.09760904 4.983159 0.034517230
6: 0.11868415 0.39541708 67.34149 0.41009950 8.383524 0.099192594
7: 0.28663647 0.08719229 139.72736 0.35239330 6.735646 0.056862360
errorMessage
1: Evaluation error: Unable to access object (object is from previous session and is now invalid).
2: Evaluation error: Unable to access object (object is from previous session and is now invalid).
3: Evaluation error: Unable to access object (object is from previous session and is now invalid).
4: Evaluation error: Unable to access object (object is from previous session and is now invalid).
5: Evaluation error: Unable to access object (object is from previous session and is now invalid).
6: Evaluation error: Unable to access object (object is from previous session and is now invalid).
7: Evaluation error: Unable to access object (object is from previous session and is now invalid).
Error in bayesOpt(FUN = MLP_model_to_tune, bounds = bounds, initPoints = 7, :
Errors encountered in initialization are listed above.
Do you have an idea about what can cause this error ?
Thanks a lot,
David
Dear all,
I have a script that worked well in RStudio on windows and RStudio server on a linux HPC. It tunes a multi-layer perceptron with ParBayesianOptimization. I tried to make it work in parallel, this way:
library(doParallel)cl <- makeCluster(2)registerDoParallel(cl)clusterExport(cl,c('OUT','X_trII', 'y_trII',"Xtr_val_data","y_trn_val_data",'my_pearson_corr','my_negative_pearson_corr','accuracy_metric'))clusterEvalQ(cl,expr= { library(keras) ; library(reticulate)})optObj <- bayesOpt(FUN = MLP_model_to_tune,bounds = bounds,initPoints = 7,iters.n = 4,iters.k = 2,kappa=2.56,eps=0,parallel=TRUE,errorHandling = "continue",plotProgress = FALSE, verbose = 1)stopCluster(cl)registerDoSEQ()but it fails:
Do you have an idea about what can cause this error ?
Thanks a lot,
David