Hi. I want to use the x-validation loop with a neronal network.

I figured out that the mean squared errors are calculated in the X-Aggregator node as followed (directly copied from the code):

for (DataRow row : in) {

RowKey key = row.getKey();

DoubleValue target = (DoubleValue)row.getCell(targetColIndex);

DoubleValue predict = (DoubleValue)row.getCell(predictColIndex);

double d = (target.getDoubleValue() - predict.getDoubleValue());

errorSum += d * d;

r++;

m_predictionTable.addRowToTable(row);

subExec.setProgress(r / (double)rowCount, "Calculating output "

+ r + "/" + rowCount + " (\"" + key + "\")");

subExec.checkCanceled();

}

errorSum = Math.sqrt(errorSum);

As I see it it is wrong. Isn't the mean sqaure error?:

mean(y(expected)-y(predicted)) * (y(expected)-y(predicted)) )

Here they calculate:

(sqrt(sum of all(y(expected)-y(predicted)) * (y(expected)-y(predicted)) )) ) /number of datasets

or is there a sense? First I thought it is the root mean square error.

But the in the root mean square error the root is calculated after the mean.