2.5. Additional Settings

Part of the network setup are settings that help how much efficiently the engine trains. This would be the percentage of randomness, the number of sessions, and for linear problems, the linear bracket.

The NumSessions is an int type that determines how many times the RLM trains. The more sessions, the more opportunities for the engine to learn and apply its learning.

rlmNet.NumSessions = 50;

Shown above is how to set up the NumSessions to 50 sessions.

The Randomness Property, defined with its min and max, dictates how much randomness is applied on the training of the engine. As it moves forward to more sessions, the randomness depreciates which allows the AI to apply its learning more as opposed to learning something new until it reaches its set EndRandomness.

rlmNet.StartRandomness = 100;
rlmNet.EndRandomness = 0;

Here we see the engine is set to start at 100% randomness, allowing it to have no limitations in trying different things, of which will depreciate as it moves further through its sessions until it reaches the last session which then its Randomness will be 0, making it only apply its learning.

Another setting is the Linear Bracket used for Linear Inputs. In summary, this allows you to work with Linear Values or multiple inputs.

rlmNet.MaxLinearBracket = 15;
rlmNet.MinLinearBracket = 3;

There is no perfect number setting for the Linear Bracket, it’s more of a trial and error and check what setting allows the engine to learn more efficiently.