This neural network is set up for use with 4 input nodes, and a single output value.

In order to use this neural network, first create a training set.  Create a list of normalized values between -1 and 1 (exclusive), along with a classification value.  close to 1 for positive (target) and close to -1 for negative (non-target).  Save this file as heimdall_train.csv

At the command prompt, type "java trainNet"

The net will then train to 10%.  If it does not, simply break the execution by pressing CTRL-C.  Change the input values to use different features.

Upon completion of training, create your testing set.  This contains a textual label, and the 4 features you wish to test.  Save this file as heimdall_test.csv

At the command prompt, type "java runNet"

The results will be displayed both to the screen, and saved in a file called heimdall_results.dat, which is editable by any regular text editor.

In order to modify the neural net architecture, open the trainNet.java and runNet.java files, and alter accordingly.  It is strongly recommended that you either backup the original files, or do not modify them at all.

For a sample file for both training and testing, copy the heimdall_train32.csv and heimdall_test32.csv to the names mentioned above and run the net.  This was our data set.  If you'd like to play around with our testing set, the images we used are included in the Images directory.

The remaining testing and training sets are included in the Data directory.

-=Heimdall's Eyes
Geoffrey Fang
Erica Liszewski

May, 2005