Add Prediction Input Dialog

The Add Prediction Input Dialog allows you to change the preprocessing associated with a prediction input which is being added.

&  For help with predictions, see Predicting and Modeling Financial Data.

Selecting Prediction Input Preprocessing

The following values are displayed on this dialog.

p    Field Name

This is the name of the field used for this input.

p    Data Series / Group

This is the data series or group which the field belongs to. In most cases, this will be the same data series or group in which the prediction is being created.

¨    Preprocessing

This value specifies the type of preprocessing that should be used on the selected field. Preprocessing normally defaults based on the type of value contained in the input field.

·      Percent Change

This option uses the percent change in the values from the previous bar as the input. It is typically used for continuous, positive, non-zero values that do not have a constant range, such as price information.

·      Change

This option uses the change in the values from the previous bar as the input. It is typically used for continuous values that may not have a constant range, such as some momentum indices.

·      None

This option uses the actual value as the input. It is typically used for discrete values that have a constant range, such as entry/exit signals.

·      Conditional

This option causes this value not to be used as an input to the prediction. Instead, it is used to determine which bars are included for training. It the value is non-zero, the bar is included. If the value is zero, it is skipped.

Conditional inputs are useful for controlling what information is sent to the neural network for training. For example, given a field that is zero when prices are flat and non-zero when prices are trending, this field could be used as a conditional input to only train the neural network on data from when the market is trending.

Ä    Note: Models with conditional inputs return values for every bar. Conditional inputs only control which data the neural network is trained with. Therefore, models with conditional inputs should typically be traded as part of a larger system which controls when to trade the signal it generates.

Ä    Note: The type of topology should be considered when using conditional inputs. Model with memory elements will not train as effectively with lots of gaps in the data.

Ä    Note: If multiple conditional inputs are specified, all of the values must be non-zero for the bar to be included for training.

Ä    Note: Change and percent change preprocessing are done relative to the previous bar of the prediction data series. In most cases, this is the change from the previous bar of the input. However, when using data of other periodicities, it will be the change from the value at the previous bar of the prediction.

&  For additional help with working with multiple periodicities, see Understanding Periodicity.

What Do I Do Next?

If you would like to add this field as an input, select a different preprocessing and press OK. If you would not like to add this input again, press Cancel.

Ä    Note: You would not typically want to add the same input twice with the same preprocessing. This is presenting the same information to the neural network multiple times and does not aid in it is training.

How Did I Get Here?

The Add Prediction Input Dialog will appear if you attempt to add an input to a prediction which has already been added and has its default preprocessing on the Predict a Value Wizard: Select Inputs page and the Modify Field Dialog: Prediction Inputs page