Predict a Value Wizard: Select Options

The Predict a Value Wizard allows you to create a new prediction field in a data series or in all data series in a portfolio or group. This page allows you to modify how the prediction will be trained and/or optimized.

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

Selecting How the Model Should Be Trained

Neural networks require a training phase to "learn" the relationship between the input values and the desired output values. Once this relationship is learned, an output value can be predicted based on the current input values. Each training phase takes some amount of time based on the amount of data being used for training and the amount of new information being gained in each iteration of the training process.

Genetic optimization can be used to try different sets of inputs and different settings in the neural network. This can be useful for finding the most informative set of inputs and/or the best neural network settings to extract information from a set of inputs. Extensive genetic optimization can take a significant amount of time since it can perform the neural network training phase several hundred times.

·      Train without optimizing the inputs (quick).

This option causes the neural networks to be trained without optimization. The currently enabled inputs and current neural network settings will be used for the training.

Ä    Note: This option is not available if one or more optimizable functions have been added as inputs.

·      Briefly optimize the inputs and settings.

This option uses genetic optimization to try some different sets of inputs and neural network settings. Using the defaults, it will try training the neural network with about 100 different combinations of settings with a maximum time of 15 minutes.

·      Extensively optimize the inputs and settings (slow).

This option uses genetic optimization to try many different sets of inputs and neural network settings. Using the defaults, it will try training the neural network with about 2000 different combinations of settings with a maximum time of 5 hours.

Ä    Note: The primary difference between "brief" and "extensive" optimization is the amount of time that will be spent trying new settings.

p    Optimization Settings…

This button displays the Training Optimization Settings Dialog, where you can select the neural network settings to optimize along with the criteria to use for evaluating each test.

p    Training Settings…

This button displays the Modify Training Settings Dialog, where you can modify the settings used for training the neural network.

Selecting a Data Range for a Prediction

The data range associated with a prediction is used for several different things. Part of the data is used for training the neural network model. Part of the data is used as a validation step to make sure the training is improving the results. And, part of the data is left out of the training process to be used for only for testing. The ranges are also used for checking the results of the optimization.

¨    What data should be used to train this model?

This setting displays the current data range being used for this prediction. It can be used to select from typical settings for the training range. Pressing the Training Range… button will display the Modify Optimization Range Dialog, which allows greater control over the date range.

By default, TradingSolutions will use the last five years of data for these tasks for end-of-day data series. For intraday data series, one week of data is used for each minute of data in each bar. These defaults can be adjusted on the Modify Default Range Dialog.

Ä    Note: For predictions being defined for an entire group or predictions that use external data as inputs, the data range is limited to the common date range for all of the data being used. For example, if the group for which you are defining a prediction has a data series with only one year of data, the prediction will only be able to use one year of data, even if the other data series in the group all have more data.

You can increase a limited data range by either importing more data into the shorter data series, or removing the data series from the prediction. You can quickly determine the first date available in your data series by adding the first date as a display field in the Portfolio View.

&  For help with display fields, see Using Display Fields.

There are no definitive rules for how much data should be used for training a prediction. Including more data is useful for improving results since it provides additional examples for training. However, older financial data may have been produced under different market conditions than more recent data, causing the training to focus on outdated or overly generalized principles.

Other Options Available From This Page

þ    View or modify the neural network model for this prediction.

By default, TradingSolutions will select an appropriate neural network for the data you are predicted. If you would like to view or modify these settings, select this option. This will cause the Configure Model page to be displayed when you press the Next button.

What Do I Do Next?

After reviewing these options, press the Next button to continue. This will display the Create Prediction page unless you select View or modify the neural network model for this prediction, in which case it will display the Configure Model page. If you would like to exit the Wizard without creating a new field, press the Cancel button. If you would like to change the inputs or desired output, press the Back button to return to the Select Inputs page.

How Did I Get Here?

The Select Prediction Model page appears after the Select Inputs page in the Predict a Value Wizard.