How to define inputs and targets for neural network pattern recognition?

Hi Everyone,

I am trying to use the neural network pattern recognition to analyze data regarding motion capture. In the examples and previous questions I have looked at, it specified that the inputs must have certain attributes. However, since it is motion, there are no specific attributes that (I think) can be used. Instead, I am trying to use this program to understand patterns in the motion and then assign 9 varying classifications of the motion. The motion is collected from each subject in 9 different angles of tilt. I want to use the program to be able to predict angle based on unclassified or classified motion data. Many trials have already been run at each of the angles of tilt, so there is plenty of motion data to input already. Motion data is organized into matrices: the rows are x, y, z (1, 2, 3) position for each patient, and then concatenated vertically, and the columns are represented by each time stamp (collected 30 Hz). Is this possible to do this using this program? What would I choose as my input and target?

Thank you!

Answer 1

[Greg Heath]

Choose a sufficiently large random sample of motion data input columns and the corresponding class index (1-9). The corresponding target matrix is obtained from

 target = full(ind2vec(classindices))

For documentation examples use the commands

 help patternnet
 doc patternnet

For other examples search BOTH the NEWSGROUP and ANSWERS using

 greg patternnet

Hope this helps

Thank you for formally accepting my answer

Greg