On Linear Separation

This assesses whether McCulloch Pitt Cells (MCP), with weights W0,W1,W2, have learnt a selected data set. One cell is used for each of (up to 3) outputs in the set.

You can try various data sets (inputs x1 and x2) plotted on the feature space. Lines are drawn (based on W0 + W1 x1 + w2 x2 = 0.5), you assess if they separate the classes, and if not try to get the cells to learn the data.

Select a data set and see it plotted. Then initialise W0..2 using the Load Random Wts button, or enter your own values (separated by spaces) and press Load from Wts. One line is drawn (if visible) for each MCP cell - and you can see if the class(es) are separated.

If not separated, press Learn and the perceptron rule will change the weights and lines are redrawn. A graph shows how the sum of errors changes. The Tidy button replots the feature space with the last lines.

If you left mouse click in the feature space, the associated x1,x2 position is reported.