Layers not supported by the neural network will get a weight=0.5.
This is to support some corner cases where we ask the TFCSPredictExtrapWeights class for an extrapolation weight for a non-supported layer.
@mfauccig @cyoung