This screen shows the results of your prediction during training and after the neural networks are finished training. For more information see Neural Network – Results Discussion

If at any time during training, you want to abort training, press the Abort Training button. After training has been aborted, you may do any of the following:

• Press the Retrain button to restart training of the neural networks.
• Press the Back button to modify the neural network output, input list, or training criteria.
• Press the Cancel button to exit the prediction wizard without saving any results.
• Press the minimize icon at the top right corner of the wizard to minimize the entire NeuroShell Trader application. Doing so will allow you to work with other programs during training. Note that using the Show Desktop Windows task bar icon, typically located next to the Windows Start button, will also minimize the NeuroShell Trader application.

(DayTrader Only) During market hours, optimization and training may be much slower if processor time is being utilized to update charts with real-time data. Use the “Don’t update chart while optimizing/training” check box to turn off the real-time updating chart during optimization/training.

After training of all the network has finished, you may do any of the following:

• Sort the results list by clicking on the column header by which the list should be sorted. Clicking once will sort the list in ascending order based upon the values in that column. Clicking a column header a second time will sort the list in descending order based upon the values in that column.
• Press the Retrain button to restart training of the neural networks.

If you have new data since you last trained your prediction you will be given the option to

Carry forward your prediction or retrain/optimize with the most recent data.

If you are optimizing, you will be given the choice to optimize starting at the last optimal settings, reoptimize starting from scratch, or simply retrain with the last optimal settings (no optimization occurs).

• Press the Finished button to end the wizard and add the prediction to your chart
• Press the Back button to modify the neural network output, input list, or training criteria.
• Press the Cancel button to exit the prediction wizard without saving any results.
• Press the Prediction Analysis button to view detailed analysis of the selected prediction.
• Press the Save as Template button to save the prediction as a new prediction template. Saving the prediction as a template allows you to apply the same prediction to different charts without having to recreate the prediction from scratch.
• Press the Print button to print a copy of the prediction.

The columns and statistics that are displayed can be customized in the Results Screen Columns section of the Prediction Wizard Interface Options. Press the Options button to change the Prediction Wizard Interface Options. For more information see Prediction Wizard Interface Options.
Notes:

• When the prediction is added to the chart and either long and/or short positions are allowed for calculating trading statistics, a price line will be displayed with annotated trades for the prediction. The price line will be split into gray, orange, and/or green sections to correspond to the data sets used to evaluate the prediction as configured on the date tab of the prediction parameters.
• When the prediction is added to the chart and no trading positions has been selected, two lines will be displayed. The blue line is the actual value of the output. The other line is the actual predicted value for the prediction and is split into gray, orange, and/or green sections to correspond to the data sets used to evaluate the prediction as configured on the date tab of the prediction parameters.
• If missing data prevents the prediction from being made (i.e. you are using the Japanese Nikki Index as an input, but it’s a Japanese Holiday, so there is no prediction made) the prediction will not generate any trading signals. Therefore, a trading system based upon the prediction will remain in the same position it was prior to the missing data. Hence, the prediction’s profit statistics are based on not trading on days with missing data.
• During optimization the objective function is calculated at exactly the Number of hidden nodes during optimization. During training the objective function is calculated for 0 hidden nodes to the Maximum number of hidden nodes during training, and the number of hidden nodes with the best results is used. This means that even if the Number of hidden nodes during optimization is exactly the same as the Maximum number of hidden nodes during training, the training results may be better than the optimization results on the same training set.
• NeuroShell Trader defaults to displaying an annualized percent return in the results summary (unless you changed the optimization objective function). If you are using a small evaluation period, keep in mind that the percentage shown is an annualized return. For instance, if the NeuroShell Trader displays an annualized return of 1000% across a one week evaluation, the actual return was approximately 20% for the week. (DayTrader Only) In the NeuroShell DayTrader, the difference becomes even greater. For example, a 1000% annualized return across a one day evaluation is approximately a 4% return for the day.
• If you have selected to insert the prediction data series instead of the prediction signal data series (Customizing Your Environment), the start and end dates shown on the graph for the training and evaluation periods will be offset by the number of bars into the future you are predicting. This is because the data for your inputs start on the date shown in the wizard but the predicted value that you are charting starts the number of bars into the future beyond that start date.
• During Optimization, if distributed optimization or network optimization is being performed, the label beside the progress percent bar will also display the number of threads/cores being utilized by the optimization. To see details of how the optimization is being distributed, double click on the label showing the number of threads/cores being utilized to display the Distributed Processing Control Panel. The Distributed Processing Control Panel shows the computer, individual count, total %, average time, last count, last % and last time for each processor. Distributed optimization can be turned on or off using the Distributed Optimization tab of the Options Dialog accessed from the Tools menu in the main NeuroShell Trader menus. For more information about distributed processing see Options – Distributed Optimization
• Note that predictions based upon data in the Mutual Fund or Money Market Fund category will have fill prices on the next close instead of the next open. This is done to better simulate actual Mutual Fund orders which are filled at the funds posted closing price the next day.
• It should be understood that trading can result in losses, and that past performance of a trading system is no guarantee of future performance.

Topics of Interest:
What are Neural Networks?
Neural Network – Results Discussion
Neural Network – Network Criteria Discussion
Troubleshooting Your Model – What to Do if You Feel You Haven’t Been Successful
Using Predictions
Error Objective Functions
Trading Objective Functions