Classify2 has 2 inputs and makes a "Buy" or "Sell" classification for the time Lookahead bars into the future. It makes the classification by examining the most recent Lookback bars for which there is a known Actual value Lookahead bars forward. These Lookback bars are called the training set. The inputs of the current bar are compared to the inputs of the bars in the training set. An output signal (classification) is generated as a very sophisticated weighted average of the Actual "Buy/Sell" indications given in the values of bars in the training set which have inputs similar to the current bar.

The Classifier Nets don't actually read or produce the strings like "Buy" and "Hold". You use positive numbers in the Actual for one category such as "Buy", and zero or negative numbers for the other category such as "Sell". The predicted output will be a number between -1 (strong probability of sell) and 1 (strong probability of buy). Numbers close to zero could be considered "Hold" (YOU would decide how close to zero a prediction should be to be considered a "Hold").

This Net has the following parameters:

Lookback - the number of bars in the training set. The training set contains the known patterns in the inputs. The training set starts at the current bar (the bar on which the indicator will be plotted) and goes back the number of bars specified in the Lookback parameter minus 1. If Lookback is 100, there will be a total of 100 known patterns in the training set. Remember that this Lookback window moves forward with the indicator output. The neural net is essentially "retraining" with each new bar.

Lookahead - the number of bars ahead for which to make the classification. If Lookahead is 2, then the Actual value two bars ahead of the current bar will be the value the Net is trained to classify. Suppose the Actual value is the momentum of the close. Then we are telling the net to classify the momentum of the close 2 days ahead. The classification of the momentum will show up as the output of the indicator on the current bar.

Actual - this is the time series you want the Net to learn to classify, e.g., momentum of the close.

Input1, Input2 - each of these inputs is a series you want to feed the Net as an input to its classification. The known patterns are found in the inputs. The known pattern for a particular bar consists of the values in the inputs at that particular bar.

Contrib1, Contrib2 - these are the contribution values for the inputs with the corresponding number. Contrib1 is the weighting you want the net to assign to input 1. The contribution is a single real number (not a time series) from zero (meaning ignore this input) to 100 (which means this input is very, very important.) Numbers greater than 100 may be used, but will probably have little more effect than 100. You can use numbers less than 1, like .6, if you want.

Note - The defaults for the parameters of the custom indicators do not make any sense in the case of the Classifier Nets. The default Actual (to be predicted) is the Close, which is not positive and negative (or zero) as the Actual should be for Classifier Nets. However, the NeuroShell Trader does not allow indicators like momentum as the default parameter, so we just used close.