FishML
This was a super simple little program I wrote to interface with code.org's introduction to AI. This program will use user input to train the neural network, and once its trained it can automatically complete the code.org process of training their 'AI' interface. (https://studio.code.org/s/aiml-2021/lessons/1/levels/1M, Intro to computer science principles, AI/ML, lesson 1)
There are two different progams: BW, which converts to black and white and only identifies outlines for the fish/not a fish excercize, and the Characteriostics versions, which lets you choose a characteristic (Circular, Rectangular, Triangular, Blue, Red, Green, Yellow) to filter for
Keybinds
| Key | Action |
|---|---|
| 9 | Begin the guess and click automatically |
| l | Change key context mode: when it's local, it will only accept events while the image window is focused. When it's not, all key events will be captured. This is useful because when it clicks, it refocuses, making repeated actions hard with local key events. Note that the automatic clicker automatically toggles this to true, otherwise it can be very hard to re-focus and turn it back off |
| 0/space | Guess and click, 0 provided because space sometimes interacts weirdly |
| g | Guess and print results |
| t,T | Train (Shift lets you enter an amount of iterations, default is 5) |
| r,R | Reset then train |
| k | Incapacitate - prevents it from being able to click at all for debugging, BW ONLY |
| a | Save image as guessed |
| s | Save as |
| d,f | Save as not fish or fish resp. and click the button to advance, BW ONLY |
| x | Execute - opens a text prompt where you can enter interactive python code for debugging |
| q | Quit |
| Extra for setup: | |
| m,b,n | Used to save screen locations of the image, left button, and right button resp. |
| w | Toggle screen scaling between 1 and 2 |
