Introduction to Neural Networks for C#, Session 10

Session Material

Predictive, or temporal, neural networks are used to attempt to predict the future. They do this by trying to locate patterns in data that may give some clue to the future values. To do this you need to select a "window" of input data to the neural network, and then decide how far the neural network is to predict into the future. For example, consider the number sequence between 1 and 10.

1
2
3
4
5
6
7
8
9
10


Now you would like to attempt to train the neural network to predict using this data. Consider a window size of 4. We will have four input neurons that will attempt to predict the fifth time element. For example, we might like the neural network to behave like this:

Input-1: 1
Input-2: 2
Input-3: 3
Input-4: 4
Output-1: 5

If the network were to behave like this, it would have predicted 5. However, you would not want to feed in real numbers such as this. No more than you would want to feed in real stock prices for a financial prediction. You should feed in percentage moves. That way the neural network can predict just as easily if we were at 5 as 5000. Using percents, the following inputs and output would be used.

Input-1: 1.0
Input-2: 0.5
Input-3: 0.66
Input-4: 0.75
Output-1: 0.8

Now the inputs are in an acceptable range for the neural network, and they are relative.

To demonstrate how to predict the obvious, the book shows how to predict the sine wave. The next chapter will show how to create a basic S&P 500 prediction program.

Videos for this Session

Video	Title
	Introduction to Neural Networks for C#(Class 10/16, Part 1/3)
	Introduction to Neural Networks for C#(Class 10/16, Part 2/3)
	Introduction to Neural Networks for C#(Class 10/16, Part 3/3)