Does SPSS Modeler automatically scale the range (continuous) fields in the Neural Network Node?
Hi,
I've got a question regarding the Neural Network Node in the SPSS Modeler (16).
Does the SPSS Modeler automatically transform the values of range (continuous) fields so that they all have the same scale before using them as inputs in the network? If yes, which transformation is used? And are the outputs also automatically rescaled to the original scale?
I couldn't find anything about it regarding the new version of the Neural Network since Modeler version 16. According to the SPSS Modeler algorithms guide, until version 15 the range fields were scaled to values between 0 and 1, which is correct because a logistic activation function were used. But since version 16 a tangens hyperbolicus activation function is used, so the values would have to be scaled to values between -1 and 1. And that's why I'm asking this question.
Many thanks in advance for your help!
Best regards, Sam
You can export the model as PMML and check the resulting file.
I believe that the new Neural Network node in SPSS Modeler 16 behaves like neural network in SPSS Statistics, where default rescaling is "Standardized". Standardized: Subtract the mean and divide by the standard deviation. In SPSS Modeler, there is no setting to change it. Rescaling is applied to continuous inputs and targets. Of course, the outputs are automatically rescaled to the original scale.
Result of hyperbolic tangent function is between -1 and 1. Input to the function doesn't have to be between -1 and 1.
The new Neural network node is introduced in version 14. Version 15 Algorithms Guide contains algorithms of the old and the new Neural Network nodes.