In this thread I am going to share some experience obtained while playing with the Encog framework for neural networks support.
First of all I created a stand-alone prototype that is able to import trade signals with the respective outcome from a file, and it does successfully build a neural network.
Then i import a second file with trade signals without their outcome form another period and was able to correctly qualify over +80% of signals, which could help to significantly improve my profitable strategy, if 80 of the not working signals are removed, i'll be massively in the green ? (Still a replay and a live walk forward will have to prove that).
I have a stand alone prototype that seems pretty successful in being able to score trading signals.
I worked on moving the code inside a strategy, while the strategy does compile.
It gives me a failure upon launch of the strategy, Encog library or some of the dependencies can not be loaded.
While some folks have been able to play with the .net 4 backward compatibility, i'll go for a clean separation and create a stand alone AI module outside NT. It allows me a better path for scalability and possible externalization of the neural networks for multiple trading computer, update the network. etc... step by step about that later
My next plan will be to create a completely stand alone module that is able to be invoked from withing NT, making it dot net version independent.
I will first think and test a few paths, most likely a socket connection integration seems the way forward.
Any folks any ideas, feel free to drop a not or advice