SMUnlimited / AMAI

Advanced Melee Artificial Intelligence Mod For Warcraft 3
Other
210 stars 34 forks source link

There is still weak feedback intensity, especially in the initial stage #440

Open jzy-chitong56 opened 1 day ago

jzy-chitong56 commented 1 day ago

As a benchmark in the industry, many people believe that your strength must be able to defeat real people and most other AI. Once someone easily wins, they will feel a sense of loss


The problem this time seems to be focused on the slow development of the military in the early stages -- About 5 minutes, 30 food, difficult

Of course, the feedback also mentioned that the intensity has improved in the later stage, as the expansion has become more active and there are resources to train the troops

SMUnlimited commented 23 hours ago

Enabled and fixed BR forgot about that logic https://github.com/SMUnlimited/AMAI/commit/e97ed38d809bd19fed99f49f0e32a8a9e30197f8

The early game actually related to players upgrading too early before training units. Thats because it ignores the strategy build order if it can't build everything its told to so tries to build other things. For tier buildings i can make sure it obeys the strategy entirely but then we have strategies where it doesn't upgrade the town hall for ages resulting in players never upgrading, this is how it used to be.

jzy-chitong56 commented 23 hours ago

set br_rush = GetLocationCreepStrength(GetUnitX(rushcreep_target), GetUnitY(rushcreep_target), 510) <= ai_strength * 1.4 // tree should have 0.4

  else
    call AttackMoveKill(rushcreep_target)
    call ReformUntilTargetDeadAM(rushcreep_target, true, false)
    call Chat(C_Done)
    call SleepInCombatAM()
    call Trace("===BR Finished , no tree ===")

Strongly recommend retaining these two logics

SMUnlimited commented 22 hours ago

True added the first part, second part not required as it will now continue and just use normal creeping if the tree is not found.

jzy-chitong56 commented 15 hours ago

For a feature that only runs once, too many global variables are used here, which is not very good At the same time, you also improved the ownstrength calculation. Actually, there is no need to count the strength of this tree, as it becomes more distorted in the later stages