R: Setting role of prediction does not work

fryasdf

Hi.

I am trying to run an R script to compute a model and then verify the model. (All of this inside a [bigger] RM process) Now I want to be compliant with the data format in RapidMiner as I want to use the nice evaluation operators, etc, i.e. I want the role of the prediction to be, in fact, prediction :-)

This is my R-version of 'apply model': (offload is the variable Im trying to predict)

rm_main = function(model, data)
{
   final = data[, .(offload)]
   data = data[, offload := NULL]
   dataAsMatrix = as.matrix(data)

   # apply the model and build a prediction
   result <-predict(model, dataAsMatrix)

   # add the prediction to the example set
   final$prediction <- result
   final$prediction = (final$prediction >= 0.5)
   
   # update the meta data
   metaData$final$prediction <<- list(type="binomial", role="prediction")
   
   return(final)
}

This however does not work: It simply does not assign the role 'prediction' to the column 'prediction' in the data table 'final'... (without complains, without errors).

Running this, in turn, does work:

rm_main = function(model, data)
{
   label = data[, .(offload)]
   data = data[, offload := NULL]
   dataAsMatrix = as.matrix(data)
  
   # apply the model and build a prediction
   result <-predict(model, dataAsMatrix)

   # add the prediction to the example set
   data$prediction <- result
   data$prediction = (data$prediction >= 0.5)

   data = data[, offload := label]

   data = data[, .(offload, prediction)]
   # update the meta data
   metaData$data$prediction <<- list(type="binomial", role="prediction")
   
   return(data)
}

In short words: when I create a new, fresh data table and want to set the role then it does not work. When I continue to use the data table given, then it works... Is this a bug?

Regards,

Fabian Werner,
Germany