Hey all, I placed #44 last year, and I just submitted my first 2 bots (2 (not 1) because I saw an immediate issue and addressed it). Submission #2 is nothing to be proud of yet, but it’s one or two steps away from something I suspect could hit gold or silver tier. I’ll get to that, no worries.
Anyway, last year’s competition IMO was a nearly perfect example of “easy to learn, hard to master.” This year, I spent an inordinate amount of time getting butt stupid pathfinding to work just to submit my first bot. This year, the game itself is simpler, but the emergent complexity of its mechanics is more complex IMO and I wonder if this is off-putting to beginners here. In comparison to last year, I believe I hit silver tier in the first 24 hours (correct me if I’m wrong) and I flirted with platinum after I eventually hit gold.
Second, my original plan this year was to build an RL client, but the limitations on memory and the deprecated nature of GPU clients dissuaded me from doing so. I still intend to build a value function from a neural network that looks at games (I’ve downloaded many days of gameplay to do so), but a lot of my initial plans to brute force this have been scrapped. Why a value function? Well, all the post-mortems last year indicated the top players had value functions for the board. I suspect if my client had had one, it might have done quite a bit better (it didn’t).