What is the exact syntax for creating NN that process a generic input into a generic output? There are lot of tutorials on images but none tutorials covering my use case and no tutorials covering the Transform and Pipeline objects general architetture I can use to infer the syntax.
I have some signals in a circuit. I observe tension and current for a long time providing a varying range of input combinations.. and I measure the output in "critical points". Then I record these signals for hours. The signals (input and output) are horizontal in the same "graph". I cut the hours registration into vertical slices/windows of few seconds. Each slice is basically
public class Data{
public float[] Input;
public float[] Output;
}
I want to train the Deep NN with 90% of the slices (of course sorted randomly) and test It with 10% of It
The slices are already in a bag I read with
LoadFromEnumerable
I found it is not possible to toy around with ML.Net fine tuning parameters like number of neurons in layers. Yes you could eventually import a ONNX model, but you need then to tune the model elsewhere. It is such a pity Microsoft made a framework that do not allows you to get hands dirty when the main goal of AI is also experimentation.
That's why I started my own framework, since all other C# frameworks are no longer maintained. Thus I'm not new to design big software systems this one will be a real math challenge. Any help is wellcome.
NeuraSharp NN framework
The rationale is that currently C# miss a maintained neural network lirary. I'm designing it to be very modular and extensible so you can experiment and play with it. Maybe in future I'll add also GPU support. There are quite a few out there, but are incomplete, no longer maintained or not easy to extend. (Or are blatantly just outdated wrappers over outdated versions of keras or tensorflow plus all the overhead of python)