Recur2 is a Recurrent Net with 2 inputs and two hidden neurons. The hidden neurons and the output neuron all use the hyperbolic tangent activation function (therefore the output signal is –1 to 1). There are two additional "recurrent" neurons that into the hidden neurons, as do the input neurons. Initially the recurrent neurons are zero. After every bar is passed completely through the neural net, however, each recurrent neuron is replaced by the sum of itself times a weight and one hidden neuron times a weight. Recurrent neurons also have the hyperbolic tangent activation function which is applied after the replacement.

scale – the number of past bars over which input scaling will take place. Recommended optimizer range is 10 to 200.

input1 to input2  -  neural network inputs. Indicators such as momentum or raw data such as open, high, low, and close are appropriate. More complicated indicators such as those in the Price Momentum category may also be used. Let the optimizer find the parameters for the indicators.

w1 to w12 – the weights in the neural network. These are similar to coefficients in regression analysis. Recommended optimizer range is –1 to 1.