The final project for CIS 112 at MCCC. This is a project that visualizes machine learning by teaching an AI how to place dots on a 3x3 grid in a specific order.
Official CIS112 (Data Struc. & Algorithm) Final Project by Shane Staret.
This is a link to my YouTube channel, featuring videos of this project on it. Here is a video that shows the program in action.
Explanation & Design Process
Essentially, this is an application that demonstrates how machine learning works. The objective is to "teach" the computer how to place the dots on the 3x3 grid in a specific manner. The specific way the we want the computer to put the dots is from the top left part of the grid, to the center left, to the bottom left, to the top center...to the bottom right. At first, the computer just randomly guesses where to place the dots. The "game" ends as soon as the computer places a dot on top of a space that already has a dot. The computer "wins" as soon as it learns how to place the dots in the correct order. Any further plays after a "win" will always result in a win since the computer finally figured out how to play the game and won't forget. As you continue to let the computer "play", it starts to learn from its past successful moves. So, if a move wasn't declared illegal, then for all successive attempts at the game, it will make that move since it is legal. The computer will always eventually place the dots in a specific order because I have the successful moves stored in a priority queue. This puts a "weight" on each grid space, and the computer learns to place a dot on each grid space from lowest weight to highest weight. The moves the computer makes each "game" are also documented and are stored within a queue. Once the game is over, the queue is unloaded and the moves are printed out in order.
I've always been extremely interested in machine learning, since the idea of acquiring knowledge seems so human to me. I always was puzzled by the idea of teaching an AI how to learn. I always thought, "How could you possibly teach something how to learn? Don't they already have to be able to learn before you can teach them anything?" Thus, I decided to try it out for myself. I wanted to visually show how a computer may learn, which is the main reason why I chose to visualize the concept by showing how a computer can learn how to place dots on a grid in a specific order. For design, I realized that the UI would be rather simple, so I just drew what I wanted before actually creating the UI through JavaFX. Planning how the AI would actually learn was a bit confusing at first and it took me about an hour to decide how I actually wanted to implement it. I finally settled on using a priority queue to weight each grid space, so that the computer would eventually discover that it should place a dot on the grid in correspondance to the space's weight.
After drawing the UI and figuring out the best way to implement the AI, I actually coded it. I got to admit, the AI was not easy to figure out how to properly implement so that the computer actually learned. The concept of machine learning and neural networks are still so extraordinary to me and I would really like to explore these ideas some more.
Thoughts & Future Plans
This project was created with the intent to show how machine learning works. Admittedly, at the moment this is a pretty dull example. Instead of showcasing just machine learning and how it works, I'd rather transform this into a "game" of sorts. Maybe change it up so that the computer learns how to play a game like checkers and then a human could actually play against it to see just how much it has learned. Overall, this project does a good job of showcasing machine learning, but it's a bit boring. So of course, I want to improve upon it. Here is a shortlist of what I will do in the future:
add a delay when the computer is "playing" so that you can visually see the dots being placed
add more grid spaces
instead of a simulation that displays a computer "learning" how to place dots on a board, teach it how to play checkers
create classes to house methods so code is more organized and readable
improve formatting of UI and add CSS to UI to make it more professional and aesthetic
possibly move UI to the web and allow other people to play against the computer that "learned"
GIFs
Showing a playthrough of multiple games until the computer is successful. I am constantly clicking the "Play Again" button and that the computer changes what it does each play based on what moves it learned were successful in the previous games. The GIF's quality is awful, but if you look closely, you can see that the pixels right below the grid keep changing because the moves are different each game. At first, they are completely random. But towards the end, the computer begins to figure out the right moves and the moves become more similar later on in the GIF.
About
The final project for CIS 112 at MCCC. This is a project that visualizes machine learning by teaching an AI how to place dots on a 3x3 grid in a specific order.
