Drag the slider to change the learning rate (maximum learning rate of 2.5, minimum of 0.1). Use the add and subtract buttons above the neural network to add/subtract layers (maximum of 8 layers, minimum of 2). Use the add and subtract buttons below the network to add nodes to the layer they correspond with (maximum of 8 nodes per layer, minimum of 2). Use the stop/start training button to play/pause training, and the reset button to reset training (re-initialize parameters). Hover over a weight (a line connecting two nodes) to view its value (red represents a negative weight value, blue a positive, with thickness corresponding with absolute value), hover over a bias (a small circle beneath a hidden or output node) to view its value (red represents a negative bias value, blue a positive, with saturation corresponding with absolute value), and hover over a node (a black/white/gray dot) to view its value (brightness corresponds with value - a value 0 or below is black, a value of 1 or above is white, and a value in between 0 and 1 is grey).
This neural network uses a special output activation function and cost function (softmax and cross-entropy, respectively), to effectively train for a classification task, such as classifying between dogs and cats, given play and sleep levels, or classifying a hand-written digit, given the pixels of an image of the digit. The actual content of this project isn't all that groundbreaking (it's not very difficult to swap out a couple formulas); the main focus is the UI, especially the NN editor. I wanted to make a UI that would be easy and pleasant to use. I think I did a pretty good job overall, but there are a few places I think I could improve. If you have any feedback, please comment.
