first things first. DO NOT PM ME ABOUT POSITIONS! Sorry if that was a little internet-loud, but this merits some emphasis; perhaps more even than I have given it. I'm /really/ busy with my current work, and you'll likely just get me annoyed for a split second (not even), then banned from any meaningful interaction online (if such a thing exists; I think it does, but then again, I'm bohemian like that).
Speaking of that, the news from the technical side of things is that I'm working on a (potentially multithreaded) system for solving enemy AI using Backward Induction-based neural nets, geometric algebra, a modified genetic algorithm using projective geometry, and smooth, dynamic computation of heuristic cost functionals (I employ a simple storage-and-interpolation scheme for heuristic, bivector-valued functional cost operators, allowing for a less time-sensitive set of computations that can be more easily reused).
blah blah blah, etc.
What you should see in game is not only a complex web of variables being introduced into the mod's main enemy classes, but also a sense that you're really fighting adaptive agents who can not only revert to training instinct, follow orders, and pass along information but also sense both danger and opportunity, learn and adapt tactics to match wits with the more skilled player, and remember how to avoid past mistakes.
The basic algorithm code is mostly finished; from here on out it's basically just deriving classes for the various agent types and integrating them with Source. Ahh, source; beneath that underdocumented mess of useless cvars and filler code, there lies a true beast. A true Snorlax, that is (Imagism which in a stroke of good humor reminds me; I still don't know my exact score on 3DMark Vantage.)first things first. DO NOT PM ME ABOUT POSITIONS! Sorry if that was a little internet-loud, but this merits some emphasis; perhaps more even than I have given it. I'm /really/ busy with my current work, and you'll likely just get me annoyed for a split second (not even), then banned from any meaningful interaction online (if such a thing exists; I think it does, but then again, I'm bohemian like that).
Speaking of that, the news from the technical side of things is that I'm working on a (potentially multithreaded) system for solving enemy AI using Backward Induction-based neural nets, geometric algebra, a modified genetic algorithm using projective geometry, and smooth, dynamic computation of heuristic cost functionals (I employ a simple storage-and-interpolation scheme for heuristic, bivector-valued functional cost operators, allowing for a less time-sensitive set of computations that can be more easily reused).
blah blah blah, etc.
What you should see in game is not only a complex web of variables being introduced into the mod's main enemy classes, but also a sense that you're really fighting adaptive agents who can not only revert to training instinct, follow orders, and pass along information but also sense both danger and opportunity, learn and adapt tactics to match wits with the more skilled player, and remember how to avoid past mistakes.
The basic algorithm code is mostly finished; from here on out it's basically just deriving classes for the various agent types and integrating them with Source. Ahh, source; beneath that underdocumented mess of useless cvars and filler code, there lies a true beast. A true Snorlax, that is (Imagism which in a stroke of good humor reminds me; I still don't know my exact score on 3DMark Vantage.)
↧