Back to help contents.
Creating a Problem
There are three ways to create a problem in the Neural Networks Applet.
- Load a Sample Graph (with data)
This is the way to load a prebuilt neural network. You can load from a set of sample problems, or load from a URL. Take note
that this expects a file with both data and graph information. This will return an error if there is
a problem.
On the menu bar, select File->Load A Sample Graph and Data. A dialog will pop up with a choice of sample problems. Choose one.
On the menu bar, select File->Load Graph and Data From URL. Type in the URL of the neural network file, and click load.
- Load Data
This is the way to load a comma-delimited data file and create a network to solve it. You can load from a set of sample
problems, or load from a URL. This will split the data equally into the training and test sets.
On the menu bar, select File->Load Sample Data. A dialog will pop up with a choice of sample problems.
Choose one.
On the menu bar, select File->Load Data From URL. Type in the URL of the data file, and click load.
- Create the Neural Net and the example set manually.
Click the Create mode tab to display the Create panel. Select the Create Node button and click on the canvas to
create at least two nodes. Then, select the Create Edge button and connect the two nodes by clicking on the parent
node and then clicking on the child node. The applet will automatically determine which nodes are input, output, or hidden.
It is required that the neural net be completely connected - thus an input node has to be connected to all the nodes of
its children's layer (whether hidden or output).
For the neural net to learn, you must give it training examples. To add training examples, click the View/Edit Examples button to open the Edit Examples dialog box. Click the
'Add New' button
on the dialog, and the applet will query you for one example. Add an example.
Solving A Problem
Click the Solve mode tab to go to Solve mode. Click Initialize Parameters to reset the learning of the applet. There
are three ways to perform backpropagation in the Neural Networks applet.
- Step through the process one step at a time by clicking the Step button.
- Step through the process 50 steps at a time by clicking the Step 50X button.
- Allow the applet to solve to a preset minimum error (.1 by default) by clicking the Step To Target Error button.
When performing multiple steps, you can stop stepping by clicking the Stop button
Feedback
To show the error plot, click the Show Plot button. To get an output given a set of inputs, click the Calculate
Output button and enter the desired inputs. The inputs given are NOT added to the test or training sets. Also, these
inputs do not affect the learning of the network in any way.
|