Post Reply
If you were logged in you would gain 3 XP for posting a reply.
You are replying to:
Support Nibbits by linking to us:
Your
StarCraft 2 Resource
More from Nibbits: WarCraft 3 Maps | StarCraft Maps | StarCraft 2 Maps | Gamer Configs
Contact | Advertising | About | Pending Changes from Users | Become an Affiliate | Database Statistics | Terms of Use | Twitter | API Documentation
I've mostly played TvP against this AI. It doesn't do very well on island maps. I can 14 CC and float to island and it will never attack the island, and it will never take an island expansion even when it makes most sense to.
Also it seems to always double gate against T but not attack right away, which is kind of negating the effects early second gate. If it's going to double gate it should at least attack your wall-in and force you to repair, backing zealots away once their shields are gone, could kill some SCVs with sentry/stalker, and back off once tanks come out.
I think it needs to adapt better to early expanding, by either expanding sooner, or putting on the pressure with drops and whatnot. Also it adapt better to early rushes by scouting proxies and taking into account your production buildings and their distance to it's base and getting appropriate defenses.
I think it could be better by doing some cheesy builds sometimes, proxies, or fake expansion into timing attack, or some low-econ builds.
It doesn't build the static D in very good places, for example it will build cannons behind a nexus so that they aren't defending the only place you could attack from on ground. And the cannons are too far from the nexus even if they were on the right side. It should also make wall-ins with gateways and such depending on it's build and the map. instead it just seems to build everything spread out randomly to no real strategic advantage.
Something I think would be cool is if it could programed to learn, by having an option for people upload game results, like replay files or something. Then a program could use all theses files and figure out why it lost and come with a counter strategy next time it seems a similar strategy. New version would contain what the AI decided to learn. Humans could also help the AI figure out why it's losing to certain builds and come up with better adjustments.