're-normalising' variables

DBeriro

Having normalised my data in GeneXproTools v5.0 prior to model evolution, I am having trouble reverting the modelled dependent variable back to its original form in a way that ensures the integrity of the model is preserved. Is this possible or do I need to export my target and modelled data and apply a suitable function? Also I presume that the normalisation is based on (x - mean) / s.d? I can't seem to find this information on the Gepsoft site. Thanks.

Dr. Candida Ferreira

Hi,

When you normalize your data within the GeneXproTools modeling environment, GeneXproTools preserves the integrity of the model throughout the modeling process in order to ensure that you, or anyone else using your model, can use the same data format that you used as input. This means that the generated model code handles all data transformations, from normalizing predictor variables to reverse-normalizing the model output (the latter applies only to regression models).

GeneXproTools supports 3 different normalization techniques: Standardization, 0/1 Normalization, and Min/Max Normalization. When in doubt about the formulas used in each case, there’s nothing better than taking a quick look at the generated code to see exactly the kind of transformation that is being made. So, yes, if you’re using Standardization you’ll see the formula above (x - mean) / s.d being used.

For more information and useful tips see the Data Normalization article on the Knowledge Base:

http://www.gepsoft.com/GeneXproTools/DataLoadingAndManipulation/DataNormalization.htm

Candida Ferreira

Howdy, Stranger!

Categories