TMultiLayerPeceptron load weights without a tree


I have a feature request for the TMultiLayerPerceptron class.
Since you can use TMultiLayerPerceptron::DumpWeights to save to a file all the neuron weights and TMultiLayerPerceptron::LoadWeights imports these weights, which completely define the network, why not have a constructor (or default constructor + LoadWeights) that uses these weights to define the network and make it ready for evaluation?

In other words, it would be nice to be able to define the network structure and nodes from the LoadWeights method so that TMultiLayerPerceptron::Evalute can be called (without knowing the number of hidden layers). This avoids the need to export as a class or to use the layout constructor.

Thank you in advance.

I see that evaluating the network through TMultiLayerPerceptron::Evaluate will be slower than exporting it as a C++ class. However, it would still be nice to have a way of importing an arbitrary network from a dump file and evaluating it.