I'm using the Normalizer node and the Normalizer (apply) to set my training and test data from two different file readers to a Min-Max-Normalization [0,1]. But the Normalizer (apply)-Node always says: "Normalized value is out of bounds."
What could be wrong?
Nothing! The data in the Normalizer has stricter bounds than the data that you normalize using the applier -- hence, when the scale/shift are applied -- the normalized data in the applier might be smaller/larger than [0,1].
Does that make sense?
I thought that the sense of using the applier is to apply the same strict bounds to another set of data without requiring to configure two normalizer nodes. but ok.
Nope, not quite. It's used to apply the exact same transformation (scale * value + offset) so that the predictors etc. will see the exact same domain...
you can try to use the z score normalization. as far as I know there are no fixed bounds, so you can apply additional datasets to the normalization afterwards, even if the bounds are greater or smaller.
So the message appears because the values delivered to the normalizer go from e.g. 2.5-3.6 and the applier gets fed with values from 2.5-4.3.