I imported a small neural network (784, 16, 16, 10) into scratch from python. Pressing the play button on the left makes the network do a test. Using the brush draws a number that can be fed to the network. The red and green colors on the canvas are the impact they have on the network. Red means they have a negative impact on the network value. Green means they have a positive. The brush sharp variable changes how long it takes for you to get a full pixel value when pressing on the canvas. Higher means it takes shorter time, lower means it takes longer. Brush bleed is the amount of bleeding it does to neighbor squares. Higher means more bleeding, lower means less bleeding. The bars on the left show how certain the network is on the number. If a bar is longer, that means the network is more certain that it's that number. The percentages are ordered from top to bottom, 0, 1, 2, 3 , 4, 5, 6, 7, 8, and 9. If the bottom bar is longer than the rest, that means the network thinks the number is 9, if the top one is longer than the rest, that means the network thinks the number is 0. The eraser tool on the bottom right corner clears the board of previous drawings.
Try drawing a number and see what the network thinks. Remember, it's very inaccurate.
