There’s nothing here as quite as exciting or fun as the last post, but this is the postmortem for how I thought everything went.
Spoiler Alert: It went pretty well.
First off, ups and downs, strengths and weaknesses. Overall, it does alright on the hard puzzles. It’s not on tier with the really advanced algorithms that pull 33% hard game winrates, but it managed an average of about 5%, which given the 2 weeks I had, I’m decent satisfied with.
It wasn’t as strong on the medium and easy puzzles as it could’ve been. It wasn’t bad, the but I only averaged about 75-80% on the easy, and 50% on the medium. There were a fair number of those that were just a corner click of 1 with no additional information that led me into a random guess loss, but my algorithm lacked 2 particular areas that had I had more time, I would’ve liked to implement: End game solutions and probability calculations.
Of course had I implemented those areas my winrate on hard would’ve increased significantly as well, but a better guessing system (probability) would’ve helped all across the board.
An end game algorithm would’ve been selective, and achieved a much smaller margin of increase in winrates, but every little bit is important.
The last to cover is that I improved the algorithm slightly from the last post. I’m going to go back and update that post and my modification to reflect the changes I made to my algorithm, but it basically was a small bump to opening up more squares using recognizable patterns (which the entire gist of my adjacency algorithm was capitalizing off common (and sometimes rarer) repeating patterns found in minesweeper).
Oh also! The actual results. Overall my algorithm placed second in the class. Alan, who I hadn’t really noticed before and didn’t showcase his algorithm the Wednesday a few of us did an initial test really surprised me with his algorithm. He successfully implemented a probability system that consistently topped my own adjacency algorithm. While both of our algorithms sat solidly in the top 2, his was definitely a slight step above mine, and is a large reason I would really look to implement a probability system if I ever did this in the future.
Overall, not a long post, just a quick update to wrap it all up.