I'm using NormalizationHelper.normalizeInputVector(), as shown in the Encog 3.3 Quick Start guide, section 2.1.5: "Using the model and denormalizing."
When called -- instead of getting an expected OneOfNNormalizer normalization like (0,1,0,0,...) -- instead I got an IndexedNormalizer normalization like (3).
With debugging I tracked this down to:
BasicNormalizationStrategy.findNormalizer() was looking for a match among this.outputNormalizers.
It was looking in the outputNormalizers because it was passed isInput=false.
And that happened because several levels down the stack NormalizationHelper.normalizeInputVector() passed false to normalizeToVector()'s argument isInput.
It looks like NormalizationHelper.normalizeInputVector() should be calling normalizeToVector() with isInput = true.
I'm using NormalizationHelper.normalizeInputVector(), as shown in the Encog 3.3 Quick Start guide, section 2.1.5: "Using the model and denormalizing."
When called -- instead of getting an expected OneOfNNormalizer normalization like (0,1,0,0,...) -- instead I got an IndexedNormalizer normalization like (3).
With debugging I tracked this down to:
isInput
.It looks like NormalizationHelper.normalizeInputVector() should be calling normalizeToVector() with
isInput
= true.