B.5 The Input Family

The Input family of components provide data to the neural network. In a few of the previous examples, we have used a signal generator to input a sine wave into the network.

Input Family palette

The three most common Input components are:

Icon	Name	Description
	Function Generator	Generates sinusoids, square waves, ramps, or user- defined functions for input to the network
	File component	Reads a variety of file types and provides the data as input to the network
	Noise	Generates uniform or Gaussian noise; often stacked on top of other components to add dither

Like the Probe family, the Input family components are stamped onto access points of components in the network. You must select the correct access point to ensure that the data is entering the network at the correct place. The most common location for an Input family component is at the preactivity access point of the first (input) Axon in the network.

The Function Generator is very simple to use. Simply stamp the Function Generator on the appropriate component, check to ensure it is on the proper access point (the access page of the Signal Generator Inspector), and select the desired waveform. You can select both the shape of the waveform (sinusoid, square wave, etc.) and the number of samples per cycle.

Function Generator Inspector

TUTORIAL EXAMPLE 11

This example uses the previous networks but allows you to experiment with the Function Generator.

NeuroSolutions Example

B.5.1 The File Component

The File component is probably the most commonly used Input component. It reads data from the file system (e.g., your computer disk) and provides that data as input to the network.

The File component

Using the File component is rather easy, but there are quite a few options. First add a File component to the Axon and ensure that it is at the proper access point. Next, open the File List page in the Inspector on the File component.

File List page of the File component Inspector

If there are files present that you would like to remove, click on the file and hit the Remove button. If you would like to add a new file to be read into the network, follow these steps:

· Click the Add button, find the data file you want to add by navigating through the file system, and then double-click the filename. The Associate File panel will appear, where you specify the file type and how it is to be used:

Associate File panel: set for testing with an ASCII-column file

· Choose a translator from the list. NeuroSolutions offers translators to cover the vast majority of file types:

NeuroSolutions provides three predefined data sets. The Training data set is the default data set and is used to train the network. Even though we haven't started training networks yet, you should always use the training set unless otherwise instructed. The Testing and Cross Validation data sets are used to test or verify that the network is learning appropriately. Since the Testing and Cross Validation sets are not used to train the network, they can be used to determine how the network will perform on data it has not seen yet. For now, always use the Training data set.

· Next, Click the Close button.

· If you have a data file with multiple columns, the Customize Column panel will be displayed next and allow you to select which columns you want to use as input. The default is to use every column of data in the file. If there are columns you would like to skip, select each column and hit the Skip button.

The Customize Column panel

• Click the Close button on this panel when you are done.
• If the data file has more columns than there are PEs in the Input Axon, NeuroSolutions will ask you if you'd like to modify the number of PEs in the Input Axon – usually you can just click "Yes"

That's all. There are quite a few steps, but it is relatively straightforward. If you add multiple files to the same File component and data set, then the data will be added together before entry into the network.

One other thing that will come in handy is the ability to normalize your files. If you open the Inspector on the File component and go to the stream page, there is a checkbox labeled Normalize. When this is checked, the data file will be normalized between the two values shown in the edit boxes left of the Normalize chechbox. This feature is used quite frequently in neural networks.

The Stream Inspector page

TUTORIAL EXAMPLE 12

This example allows the user to manipulate the file component and load data files into the network.

NeuroSolutions Example

Go to next section