Network




		Additional Information

Configure Neural Net: While the use of Wizard to setup networks is HIGHLY recommended. Sometimes you will still have trouble with training.

Manual control of network configuration is possible through this tab.

TIP: When modifying Wizard recommendations- make small changes. Adding a single neuron to hidden layer can make a huge difference in how network trains.

Inputs: Number of Inputs to Neural network. The smaller the value, the more generalized the Network

Outputs: Number of outputs to compute

Hidden Layer 1: # of neurons in hidden layer 1.

Hidden Layer 2: # of neurons in hidden layer 2.

Hidden Layer 3: # of neurons in hidden layer 3.

The values for Hidden Layers depends greatly on the type of problem you want to solve. Here are some general rules of thumb to go by:

Problem is linear (ie. if you put it on a graph, it makes a straight line)
Very few Hidden Neurons required. Psychonet should handle linear problems if enough sample data is provided. Keep the inputs as small as possible.

Problem is nonlinear and simple
Neurons required in Hidden Layer one - Should not take many neurons for simple problems however the actual number of neurons will vary depending on the number of training samples produced from your raw data.

Problem is nonlinear and complex
Neurons may be required in Hidden layer 2 - If you cannot get the Network to train adequately with hidden neurons in layer one, then start adding neurons to layer 2. This will add time to the training cycles, so use layer 2 sparingly.

Problem is EXTREMELY complicated
Neurons may be needed in Hidden Layer 3 - If all else fails you can try adding layer 3 neurons however this will dramatically increase training time. Its use is not recommended for the general public.