|
Navigation
Search General Browse
» All Files
» All Maps » Maps: Assault » Maps: Board/Card » Maps: Bounds » Maps: Defense » Maps: Micro » Maps: Movie » Maps: Puzzle » Maps: RPG » Maps: Strategy » Maps: Terrain » Maps: Others » All Mods » Mods: GRP Unit » Mods: GRP Building » Mods: GRP Misc » Mods: Portrait » Mods: Conversions » Mods: Others » All Programs » All Sounds Sunken Treasure
|
Type-Moon Arena vR99b
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||
[02:49 pm]
Oh_Man -- it doesnt need maphack. the person who stole it and is using it on the ladder is using it with maphack. all its training was done with normal fog of war[01:28 pm]
NimoStar -- Unless specifically logged and retrained. But that wasn't in the original version. It just played itself[01:28 pm]
NimoStar -- it has a static training, which is converted into weights. it doesn't actually learn from new games.[01:05 pm]
NudeRaider -- Oh_ManOh_Man shouted: No training is very computational expensivr it doesnt train in real time tht culd actually be used against it
oh but alphastar training was *very* cpu intensive.[11:31 am]
Oh_Man -- No training is very computational expensivr it doesnt train in real time tht culd actually be used against it[11:11 am]
NudeRaider -- Oh_ManOh_Man shouted: Hope that makes sense. Also artosiscast just dropped a 1 hour vid he analayses a bunch of its games. The TLDR is the maphack is responsible for a large percentage of its wins because its just hard countering zerg rushes
People keep mentioning it's a bot using reinforced self-learning (or w/e the correct term is) but I wonder, does it still actively learn now that it is deployed? How does that work? Or was it trained on like 100k games playing against itself and now training is finished and its capabilties are now static?