Perceptron will not train if yOffset not zero.

Not sure what I'm doing wrong.
[SimplePerceptronApp.zip](https://github.com/CodingTrain/Rainbow-Topics/files/3154906/SimplePerceptronApp.zip)

On the canvas, at bottom left is a "button" that you click to scroll through various parameters that can be incremented/decremented.
There's another "button" bottom/center that can "train" (one-shot), or continuously if you hold down the mouse click long enough. You can then click the "button" to stop the training.
The Plus/Minus "buttons" will inc/dec the selected parameter.

The display will show colored circles and indicate which it got wrong after end of a training cycle.
The gray color means it got that circle's category wrong.

If I leave the slope starting at 0,0, then you'll see the subsequent trainings go to zero errors.
But if you increment the yOffset, it never seems to "learn". The console output shows weights just flopping around.