I put some good work into my bot's navigation and collision avoidance logic, and merged that with a decent strategy to get into the top 16 before I took a bit of a break.
I'd like to build a CI/CD pipeline to play versions of my bots against eachother, and analyze those private replays along with the publicly available gold replays to help drive some decisions around new strategies to pursue. I think predicting outcomes on a turn-by-turn basis could help identify portions of games, or particular map styles where my bot struggles.
Right now, I can tell from human inspection that my bot is doing okay, but I've made some decisions about ship allocation for defense/production that are leading to an overly aggressive early game. It's working okay, but I think I could do better by trying to figure out how many ships I actually need to allocate to fighting.
I've got some ideas about trying to use reinforcement learning specifically on ships in combat situations, which I should be able to accomplish through some modifications to my ships' role-based behavior, but I'm not immediately in a place to take advantage of this money for that purpose.