#mainTextBox NeuroSolutions is a fully OLE-compliant server. This means that NeuroSolutions can receive control messages and data from other applications and respond by returning data. There is a close relationship between NeuroSolutions' OLE commands and its macro language. The OLE commands are often "wrapped" macro statements. The data set for this demo is identical to the discriminant function demo (within the Dynamic Link Library Demos), and consists of two classes: an inner box of four points and an outer diamond of four points.
#subtitleTextBox Building a network
#mainTextBox We have created a Visual Basic (VB) application, whose interface is displayed at right. This application sends OLE commands to NeuroSolutions. The easiest way to create the code for such an application is to create a macro within NeuroSolutions that performs the actions you want, and then use the "Copy as VB" command from the MacroWizard interface. This converts the macro statements into VB OLE commands, which then can be pasted directly into any VB-compatible application, such as Microsoft Excel and Access. Run the application to build the network, and then view the VB source code. You won't be able to advance until you build the network. Note: if the application window becomes hidden, click on the right-most task bar button.
#subtitleTextBox Training a network
#mainTextBox We have created another Visual Basic application, this time for training the network. When the "train" button is pushed, the application sends OLE commands to NeuroSolutions to set the number of epochs and the learning rate for all backprop components. It then sends a run command. When the training is finished, it requests the final mean square error from NeuroSolutions, which it then displays. Train the network, and then view the VB source code. When you are finished, close the VB application.
#subtitleTextBox Preparing to test a network
#mainTextBox Once a network has been trained, it can then be prepared for testing on new data. Such preparation would normally entail removing the backprop components to freeze the training and speed up testing, and loading in a test file at the input. All these operations can be performed through external OLE commands. For the test input, we use the OLEInput component, which allows the user to send data directly to NeuroSolutions from another application. Prepare the network for training, and then view the VB source code. You won't be able to advance until you prepare the network.
#subtitleTextBox Testing a network
#mainTextBox The OLEInput component accepts 2-dimensional arrays of single precision numbers. The 1st dimension represents the channel number, and the 2nd dimension represents the exemplar number. We have created a VB application to send input data to NeuroSolutions one exemplar at a time. Type in values for the input and hit the "test" button. The input data, a run command and a request for the probe data are all sent to NeuroSolutions. NeuroSolutions receives and prepares the input data, runs the network and returns the output, which is then displayed in the VB application. Try it, and then view the VB source code. When you are finished, close the VB application.
#subtitleTextBox Summary
#mainTextBox You have seen how NeuroSolution's OLE compatibility gives you tremendous flexibility. Once you have defined your favorite neural architectures, you can do all of your network training and testing external to NeuroSolutions. In fact, NeuroSolutions' OLE command set is so extensive that you need not ever work within NeuroSolutions itself.
