# Reversing Normalization: How?

Hi all,

I'm doing regression models (LinReg, PCA, epsilon-SVM, ..); hence it is common practice to normalize the data set. But I'd like to get the prediction in the original coordinate set.

Since normalization is just a linear transformation, this is obviously easy.... if RM reveals

Thanks for your help! Stefan

PS: the GUI of v5.0 is a big step forward - still some rough eges here and there, but now I spend my time thinking about what my data is telling me, rather than trying to figure out how to tame RM! Thanks a lot! Great job!

I'm doing regression models (LinReg, PCA, epsilon-SVM, ..); hence it is common practice to normalize the data set. But I'd like to get the prediction in the original coordinate set.

Since normalization is just a linear transformation, this is obviously easy.... if RM reveals

*what exactly*it has done.... Is there an operator which can take a preprocessor model to reverse the transformation?Thanks for your help! Stefan

PS: the GUI of v5.0 is a big step forward - still some rough eges here and there, but now I spend my time thinking about what my data is telling me, rather than trying to figure out how to tame RM! Thanks a lot! Great job!

0

## Answers

2,531Unicornsince it's a linear transformation most regression methods simply don't care if data has been normalized, but this only as a small remark.

You could take a look at the normalization model, there should be noted, which parameter it is using. But there's no operator for inverting this. At least not until now...

Thank you very much for your kind words, if you have any suggestions feel free to give any hints what we could do better

Greetings,

Sebastian

53Guruimplicitly, you said if of course:

:mostregression methods simply don't care if data has been normalizedCertainly SVM cares (which is also in line with the web page on LibSVM).

Stefan

44GuruHopefully this will help.

Also I agree the new interface in RM 5 is fantastic.

Cheers,

Cleo

2,531Unicornthanks for this kind works, too. I will tell my colleges

About the idea with joining: This would work if the normalization uses a view. Otherwise the underlying data is changed and I guess, the original example set will exactly look like the normalized...You could afterwards materialize the normalized example set, which will create a full copy in memory (or do it before and normalize it using an IOMultiplier)

Before joining you might filter out all attributes except the id and the prediction. Then you would not need to uncheck remove double attribute.

Greetings,

Sebastian