What is this?
As part of the Machine Learning class in (CIS 678) at GVSU, Camila and Roland created the neural network J.O.M.I.S. (= Just one more intelligent system ). The network creates a MLP (multi-layered perceptron) based on the input data and one hidden layer. It immediately starts training and will display the global error and the network graph below. After completion you may use the "manual verification history"-input-field to see prediction by the network with unseen or seen data.
Uploading your own
You may upload your own text-file. Please notice that the script is running in your browser and may cause performance issues.
Sample formatting for file:
Health Armor Weapon Enemies Action Good Yes No 1 Attack Good Yes No 2 Run Good Yes No 3 Run Good No Yes 1 Attack Good No Yes 2 Attack
Please note that the first line contains the names of the attributes.
Neural Network Visualization
Following graph shows the created neural network. Each node represents a neuron of the network. The nodes are grouped into input nodes (left column), hidden layer nodes (center column) and output nodes (right column). The edges that connect two neurons represents their "relationship". The thickness of the lines corresponds to the absolute value for the weight of the edges. The colors green and red indicate if the weight-value was positive (green) or negative (red).
Network- and Classification-Error
Following chart shows the global network error over epochs passed (blue) and the classification error of the verification sample (red).