| New Features |
| New Short-cut menus for performing tagging,
preprocessing and analyzing functions that require the selection of columns or rows. |
| Powerful data cleaning preprocessor -- finds missing values, specified values, and/or error codes and replaces them using interpolation, extrapolation, the column average, or a specified value. |
| New status dialog for training procedures allows the training
to be stopped at any time. |
| New option on training dialogs allows the training to automatically be stopped if the cross validation error has not improved within the specified number of epochs. |
| Genetic training -- creates more accurate models by optimizing the specified network inputs and various network parameters, such as step sizes, number of processing elements, etc. |
| New Production dataset type -- provides an easy way to apply a dataset that does not have a desired output. |
| Enhanced Usability |
| New and improved look and layout for all dialogs. |
| More intuitive menu and menu item arrangement. |
| More informative messages. |
| Labels can now contain the following characters: ", ; : `. |
| The Moving Average preprocessing function no longer requires the Analysis Toolpak to be installed and activated. |
| The Moving Average preprocessing function can now be performed on a contiguous or non-contiguous multiple column selection. |
| The Sample preprocessing function now maintains Input and Desired tags if they existed on
the original worksheet. |
| Termination options are now available for networks built using the NeuralBuilder from within NeuroSolutions for Excel. |
| The Training percentage can now be specified in the Tag Rows By Percentages dialog. |
| The New Batch dialog now provides the ability to specify the batch name and description. |
| The Batch Manager dialog now provides the ability to edit the selected batch and change the type of batch being viewed. |
| The batch type being viewed on the Run Batch dialog can now be changed. |
| For dynamic networks in trajectory mode, all training procedures ensure that the number of training exemplars is evenly divisible by the trajectory length by automatically untagging the additional training rows. |
| Cells containing formulas can now be tagged and used for training/testing. |
NeuroSolutions
