Error '3' when training a model

[Pedro Henrique Veronezi e Sá]

Hello!

I'm training a model using R, and I'm getting the following error:

Got exception 'class java.lang.ArrayIndexOutOfBoundsException', with msg '3'
java.lang.ArrayIndexOutOfBoundsException: 3
	at hex.tree.SharedTree.initial_MSE(SharedTree.java:606)
	at hex.tree.SharedTree$Driver.compute2(SharedTree.java:168)
	at water.H2O$H2OCountedCompleter.compute(H2O.java:1069)
	at jsr166y.CountedCompleter.exec(CountedCompleter.java:468)
	at jsr166y.ForkJoinTask.doExec(ForkJoinTask.java:263)
	at jsr166y.ForkJoinPool$WorkQueue.runTask(ForkJoinPool.java:974)
	at jsr166y.ForkJoinPool.runWorker(ForkJoinPool.java:1477)
	at jsr166y.ForkJoinWorkerThread.run(ForkJoinWorkerThread.java:104)


 Hide Traceback
 
 Rerun with Debug
 
Error: '3'
 
10
 
stop(m, call. = FALSE)
 
9
 
doTryCatch(return(expr), name, parentenv, handler)
 
8
 
tryCatchOne(expr, names, parentenv, handlers[[1L]])
 
7
 
tryCatchList(expr, classes, parentenv, handlers)
 
6
 
tryCatch({
    while (keepRunning) {
        myJobUrlSuffix <- paste0(.h2o.__JOBS, "/", job_key)
        rawResponse <- .h2o.doSafeGET(urlSuffix = myJobUrlSuffix) ...
 
5
 
.h2o.__waitOnJob(object@job_key)
 
4
 
h2o.getFutureModel(job)
 
3
 
.h2o.modelJob("drf", parms)
 
2
 
h2o.randomForest(x = predictors_, y = response_, training_frame = train_, 
    validation_frame = valid_, sample_rate = 0.7, ntrees = 50, 
    max_depth = 20, min_rows = 1, nbins = 100) at
 get_results_fun.R#2125
1
 
DRF_final_tier_userdefined(response_, selected_predictors_, train, 
    valid)
 

The code that I'm running is this one, it is a loop to train a series of models, one for each category that I have. This is like the final step of a series of complex models, but only here I'm getting an error.

train_valid <- as.data.frame(train_valid_)

  

  #executes the training by notches

  for (i in 1:length(notches)){

    df_train <- train_valid[which((train_valid$predict_picked == as.character(notches[[i]][1]))|(train_valid$predict_picked == as.character(notches[[i]][2]))|(train_valid$predict_picked == as.character(notches[[i]][3]))),]

    if (length(df_train[,1])>0){

      df_h2o <- as.h2o(df_train)

      #splits the dataset into train, valid, usually is 80%, 20% just used when the test data were picked already

      splits <- h2o.splitFrame(df_h2o, c(0.8))

      train  <- h2o.assign(splits[[1]], "train.hex") # 80%

      valid  <- h2o.assign(splits[[2]], "valid.hex") # 20%

      #transform all columns needed to factors

      train$rating_crude <- as.factor(train$rating_crude)

      train$rating_full <- as.factor(train$rating_full)

      train$sector <- as.factor(train$sector)

      train$Ticker <- as.factor(train$Ticker)

      valid$rating_crude <- as.factor(valid$rating_crude)

      valid$rating_full <- as.factor(valid$rating_full)

      valid$sector <- as.factor(valid$sector)

      valid$Ticker <- as.factor(valid$Ticker)

      DL_picked <- DL_gridsearch(response_,selected_predictors_,train,valid,num_models_=5,best_error_=1)

      DRF_picked <-  DRF_final_tier_userdefined(response_,selected_predictors_,train,valid)

      err_dl_model <- h2o.confusionMatrix(h2o.performance(DL_picked,valid=T))$Error[length(h2o.confusionMatrix(h2o.performance(DL_picked,valid=T)))-1]

      err_drf_model <- h2o.confusionMatrix(h2o.performance(DRF_picked,valid=T))$Error[length(h2o.confusionMatrix(h2o.performance(DRF_picked,valid=T)))-1]

      

      if ((err_dl_model <= err_drf_model)&((err_drf_model != 1)|(err_dl_model != 1))){

        assign(paste0("not",i,"_model"),DL_picked)

      }else if((err_drf_model != 1)|(err_dl_model != 1)){

        assign(paste0("not",i,"_model"),DRF_picked)

      }else{

        assign(paste0("not",i,"_model"),0)

      }

    }else{

      assign(paste0("not",i,"_model"),0)

    }

  }

Do you guys think that maybe there is a limit for number of models in the same h2o session?

I have no idea whay this is happening and neither where to find this error on the documentation, someone has had this kind of trouble before?

Thanks in advance

Pedro Veronezi

[Spencer Aiello]

What's the version of H2O that you're using?

[Pedro Henrique Veronezi e Sá]

My h2o version is>

packageVersion("h2o")
[1] ‘3.6.0.3’

and my R version is>

> R.version
               _                           
platform       x86_64-redhat-linux-gnu     
arch           x86_64                      
os             linux-gnu                   
system         x86_64, linux-gnu           
status                                     
major          3                           
minor          2.0                         
year           2015                        
month          04                          
day            16                          
svn rev        68180                       
language       R                           
version.string R version 3.2.0 (2015-04-16)
nickname       Full of Ingredients   

should I update something?