| 07-05-2004, 05:53 PM | #226 |
Great news, downloading AMAI 2.2. |
| 07-05-2004, 06:05 PM | #227 |
I must admit, that Army Tracking code you put in is the greatest thing on the face of this earth. |
| 07-05-2004, 07:15 PM | #228 |
I just tested this new version of AMAI 2.2 for my RoC game. On my favorite small map Frostsabre , i was Orc versus an Undead AMAI (ArraXy) I used the Blademaster supported by several grunts and annihilated AMAI while killing 6 times its Dreadlord in 13 minutes :) . But i noticed an important improvement on the previous AMAI about a behaviour problem i reported earlier. In the previous AMAI version, the AMAI armies went all to attack my army , mostly ignoring my hero so i was free to kill the enemy AMAI hero). It was an important problem as it was easy to defeat AMAI once his hero was killed. In AMAI 2.2, i noticed during the fights that a part of AMAI army attacked my hero, while another part was going to attack my grunts. That is a very lot more interesting despite the fact that on small maps the BladeMaster is really an hero killer (especially when you have grunt to hit your sleeping BladeMaster :) ) . I do not know if you have modifed AMAI code on this point, if it is a good addition from the ArmyTracker , or if i was just lucky to find an AMAI personnality willing to attack my hero , but it is definitively a great thing as it makes the combat less easier. I noticed that again , as in preivous versions, AMAI never upgraded his ziggurats letting me totally free to move in his base without any danger (in RoC ziggurats and black citadels really hurt). UD Blizzard AI upgraded towers are not making things easy for the player. But it is possible that my constant attack on AMAI never left any time for it to upgrade the towers. I will see in bigger maps if AMAI decide to upgrade , as it will have more time (and so money). Congratulations on AMAI 2.2 , i had a lot more fun defeating "ArraXy" than in the previous version, it is more challenging now. Thanks. |
| 07-06-2004, 01:06 PM | #229 |
Hi, Something I've been wondering about for a while - the Human AI has this code in it: // Anti rush if GetRushFactor() < 7000 then call SetBuildUnit(2, ARCANE_TOWER, 70) endif Can you give an example of a map that has a RushFactor of < 7000? The 3 main rush maps I could think of all had RushFactors greater than that. (2)Circumvention == 8441 (4)Avalanche == 9408 (4)TwistedMeadows == 8177 Thanks in advance, Goose. |
| 07-07-2004, 08:55 AM | #230 |
I just played the AMAI 2.20 and I see the the "Army Tracker" works quite well and the comp seems to bring it even harder this time while still building up it's base and teching up. Yet, unfortunately, the same problem occurs. The comp will not rush to expand. It will attack me and use it's resources to attack other creeps but not attempt to attain more resources. Tried to adjust the strength level that the comp sees but still didn't do a whole lot. The Human ai seemed to poke at one of the expansions but then left it alone to fight other creeps despite being in tier 3 with plenty of firepower and dwindling gold. Wound up beating me despite having no more gold and still had some expansions untouched (see my previous post to see the level and types of creeps I put at expansions for testing). Andy, you and Zal have done a great job but this may be the one lacking aspect lagging in the AMAI program. Is there a way to tweak this with this release? Will I and can I create a trigger or adjust the global settings to force the AI to expand? Thanks in advance. |
| 07-07-2004, 12:49 PM | #231 |
@Goose: The rush factor is computed by taking the distance to the nearest enemy start location and then adding 1000 for each unused player slot (not a very complicated calculation, can probably still be improved a lot). I guess if it is larger than 7000 on the major rush maps then that value needs to be higher there (that is Zalamander's job, I just provide the code, and I guess he currently investigates that value). @Zeta: There are 3 things that need to be true before they expand. The first is the expansion time (did you actually put that to a low value in the strategies?). The second is that they consider the expansion reachable through the pathing algorithm (they will not expand to expansions surrounded by inpassable stuff at the beginning of the map and waygates are not taken into account). The third is the strength of the army that guards the expansion. Only if he is both stronger in total and in air/anti air, then he will attack. Important there is that table I mentioned that determines how strong a creep of a certain level is considered. Also note that if there are other creeps within a certain radius they might be considered as belonging to the expansion guard army. |
| 07-07-2004, 06:56 PM | #232 |
Today, i wanted to play just for fun, i went to a mass acolyte strategy in Frostsabre versus [AMAI]Crusher (in Normal difficulty level) that was Undead too. At my big surprise , i won in 18 minutes (that was my first Mass peons ever). But when watching the replay from [AMAI]Crusher point of view i noticed 2 problems. -The first one, a bug that is new for me, as i do not remember having noticed it in previous versions. As i made a Mass Acolyte , i decided to create a not too far of the enemy base. So without surprise, [AMAI]Crusher, with his Dreadlord and ghouls attacked this position several time, but with my Dreadlord and acolytes i succeeded to stop each of his attacks and even killed him (Acolyte surround is a crazy thing :) ) Then later, at the 4th minutes and 40 seconds, AMAI began another attack, but my advanced base was ready in the front, with upgraded ziggurat towers. At the the 5th minutes , AMAI retreated to his base as my defense of the advanced base was really good. And there in his base , at the 5th minutes and 30 seconds , 5 of his own ghouls attacked his own Dreadlord and killed him !! -The second one, is an old behaviour problem already there in previous versions. Later , at 11:25 the AMAI Dreadlord and the Deathknight he made attacked a creep camp with ghouls and fiends. Then after the creeps were killed, its army went back to his base , but the Deathknight continued alone. His objective was a far away Mercenary camp.. But my advanced base was exactly in his patch, the Deathknight was wounded. He arrived at the Mercenary base at 11:57 He stayed there until 12:12 and left , going through my advanced base (again he was very wounded). But i noticed that he bought absolutely nothing, despite at the moment he arrived in the Mercenary camp AMAI had +600 gold !! Is there a logic explanation to the murder of his own Dreadlord or is this as i think a bug ? And why AMAI never bought any mercenary despite he went 2 time through my advanced base defenses , and so was nearly killed, this happened before, is this related to the fact it is a ROC Mercenary camp , and the code is trying to see if TFT mercenaries are there. ? I attach here the AMAI 2.2 Frostsabre map and the replay of this game if you want to have a look for debugging. The map must be put in ...\Warcraft III\Maps\AMAI\ It is a ROC version. |
| 07-07-2004, 06:58 PM | #233 |
And the replay of this ROC game |
| 07-07-2004, 09:39 PM | #234 | |
Quote:
TURTLE ROCK. WHEN YOU GET TO BE NEXT TO EACH OTHER.CHEERS :-) |
| 07-08-2004, 03:04 AM | #235 |
Thanks for the response Andy. Let me be specific. I have already had the times low on the strategies that I use and they are locked. They are also low on the "personalities" as well. I have modded the Twilight Ruins map and have a Razorman Medicine man(L5), 2 Centaur Impaler(L4), 2 Harpy Storm hags(L5), and Centaur Kahn (L8) at each of the center goldmines. The comp seems to come close but never turn in towards those mines when it goes creeping. I did not however realize that the distance between creeps can make a difference so I'll try that and I also changed the levels in the LevelStrengthMap file but it didn't seem to work. I am about to try again so I will post after I play and watch the replay. Thanks again for the responses though. |
| 07-08-2004, 11:19 PM | #236 |
Hi guys I've just played a funny game with the AMAI2.2. 1.Zmasher has build a townhall at the center of (4)Hellfire.w3x. 2. Terra has build 2 altars. 3. At the end of the match, 1 of Zmasher's Knights attacked 1 of its own. o_O PS I hope my english isnt 2 bad. ^_^ Here is the replay if y wanna see it. |
| 07-09-2004, 12:33 AM | #237 | ||
Quote:
Ah.. thanks for the explanation! The code in GetRushFactor() didn't make any sense to me because I guessed GetPlayers() returned the actual number of players in the game, but it sounds like it returns the maximum possible players. That would explain it :) It sounds like this code is geared towards 1v1 play (for instance a 1v1 on Turtle Rock like Xakepa suggested). However for 2v2 games where all start locations are used up, a rush is pretty likely on say (4)Avalanche. I think I might try messing around with the Arcane Tower stuff in my copy. I think I'll put the rush factor check up to 8500, if so build 1 arcane tower. Then remove all the other arcane tower references in the code. I personally only value them for stopping hero rushes. Thanks for the information :) As always, thanks for the hard work! Quote:
|
| 07-09-2004, 03:39 PM | #238 |
@boltlore: Thx for the replay. It will allow us to investigate that bugs. Number 2 is not really a bug. The AI is actually allowed to build a second altar when it has many high level heroes to revive and enough resources. I have already seen Number 3 one time. It might be a bug in the focus fire code (unless it is a hard coded problem). Number 1 is likely caused by a location being destroyed at the wrong point or not filled at all. |
| 07-09-2004, 04:10 PM | #239 |
The Chinese Chatting system works perfect in my 1.16 TFT Chn.ver. Thanks for your work! |
| 07-09-2004, 04:34 PM | #240 |
@AiAndy I have also seen problem number 3 occur about twice before even in the last version. Its unlikely its hard coded in the ai cause i never seen standard ai attack its own units. |
