Transformed Regression operator with

yzan

When I set  "transformation method"="none" in Transformed Regression operator, all the predictions have zero value. The attached example illustrates the issue (in Rapidminer 9.4).

The workaround: Do not use the operator when not necessary.

It is also possible that I interpret the "none" option badly or that I use a bad flow. If that is true, the documentation should be extended/an example flow should be provided.

Marco_Boeck

Hi,

Thank you for the report and even providing the code that was wrong! This will be fixed in the upcoming RapidMiner Studio 9.5 version.

Regards,
Marco

sgenzer

hi @yzan so thank you for this. The good (bad?) news is that this operator is rarely used any more. In fact many of us here had the same reaction: "Transformed Regression? Never heard of it!". Hence it is highly unlikely that we will add an example or extend documentation to this due to low usage. That said if you have a process (similar to the one you posted but perhaps with some sticky note explanations) that would help users, I can add it to the community repo. 

Thank you!

Scott

yzan

The bug is in

com.rapidminer.operator.learner.meta.TransformedRegressionModel.java

in section:

case NONE:
	if (zscale) {
		while (originalReader.hasNext()) {
			double functionValue = reader.next().getPredictedLabel() * stddev + mean;
			Example example = originalReader.next();
			example.setPredictedLabel(functionValue);
		}
	}
	break;

Currently, when case==NONE and zscale==False, it doesn't do anything. The corrected code could look like:

case NONE:
	while (originalReader.hasNext()) {
		double functionValue = reader.next().getPredictedLabel();
		if (zscale) {
			functionValue = functionValue * stddev + mean;
		}
		Example example = originalReader.next();
		example.setPredictedLabel(functionValue);
	}
	break;

sgenzer

cc @Marco_Boeck?

yzan

Thank you.