@Moderators & Admins: Is it ok to posts Updates here?
I have read your rules: "Don't use our forums as your personal marketing resource. Don't post referral links or blatant traffic grabs. It's fine to post a thread that links to your personal blog or an article you wrote or published on another site if the content is relevant to the discussion, but treat our forums as your home, not as a tool to drive traffic."
I guess this means, exzessive bumping is not tolerated and I do not plan to do it. But users like - LordYama and - Xapti and many others gave me a lot of hints and ideas which will take time to program.
I would like to inform these ppl if I have a "Xapti-Feature" ready.
The changelog I will post it in my first thread, using the EDIT function and using Spoiler-BB-Code.
> Definitely drop harrassment would be good
If you search for "NeoMT" you will find Alterzim. The AI will do Drops in the next few days. I worked on other harassment styles, though. I am glad to report: You will be attacked by cloaked banshees. If you have good map vision you can attack the Banshees before they cloak.
I implemented Reaper-Scouting, but Reaper Harass is still work in progress.
> Misleading thread title, as this post is about custom AI, not improvements in Blizzard's AI.
Yes and no. My AI executes openings like any other Blizzard-AI. Protoss-Blizzard-AI often builds 2 RoboFacilties at once. It still does. Improvement: It actually builds something in those Robos. Blizzard-AI builds a Stargate. Improvement: My AI will build Tempest, which the Blizzard-AI won't do.
It's not my goal to create a new GTAI. It's my goal to improve the Blizzard-AI. E. G. A big problem for all of them was: SupplyBlocks. Terran AI: My first version had a TSC (Time Supply Capped) of "7:28". I will try to reduce it below 0:59.
===
Updates 2014-06-26 General: - I will do a new Zerg-AI (Overgrowth) without Phoenix and other triggers: 7Pool and 10Pool - Note: 6Pool is more complicated than I thought, because AI will build at least 1 Drone
- Added Hidden Nexus at three different places - If FleetBeacon is present: Tempest - Random Integer: Tempest or VoidRay - Bugfix: Sometimes AI just builds Immos => Build 1 RoboBay (Collossi) - Added "Korean Macro": Reduce your Minerals to 0 or below 100
- Arcade Lobby still is broken ... Player User => None - Bugfix: Objective "Deny enemy expansions" never failed - HiddenCC: Decreased from 80% to 60% - Changed positions of SupplyDepots (Help AI) - Air Strike 2 needs testing (Raven Harass) - Added Air Strike 3: Cloaked Banshees - AI bug: Sometimes AI does not research stim => FIXED - Added Rax with TechLab (EarlyGame) - Added Air Strike 2 (Raven) - Bugfix: Forgot TestAction +2K Minerals
- Phoenix now may die. Level: "Very easy" and "Easy" - Checked some Bullies (Default vs Always) - Changed message "Which AI?" - Kill trigger: Increased Minerals (900 => 1.000) - Moved positions of Phoenix
If it doesnt already exist, i'd like to have an option to play a build on my own, and then ai to mimic what i did so i can test things like timings etc against it without having to call for a practice partner.
This would be awesome. Like the old Need for Speed games where you could race against your chost, and see if you can beat it.
I play AI too much. Especially for someone in diamond and I have spent a lot of time with AI's in both BW and WCIII
In my opinion, the most overlooked thing in AI design is that they are trying to make 1 generic AI that will work on all maps. This is not the best way to go about it! Having an AI framework for any map is of course vital, with a selection of build orders and such. But, giving it map specific variables and "regions" for it to scout, build in more often, or prioritize over others, should be done.
Think about a human player, there are areas you are going to scout more for proxys, places you are more likely to cannon rush based on which map you are on ... and blizzard should not be afraid to implement these into the AI. The old Green Tea AI would cannon rush, and I am sure it could be programmed to proxy and randomize it's build orders more than the current AI. Upping the "micro" of the AI is one way to go about it, but really giving the AI a real variety of build orders, and more if/else statements regarding scouting of early hatches, early gas etc based on scouting would be VERY interesting.
Updates 2014-07-05 General: - Working on drops (WarpPrism) - Unloading works, but Units won't attack
> I do however think this is not possible
You are right.
In my opinion, the most overlooked thing in AI design is that they are trying to make 1 generic AI that will work on all maps.
Please take a look at my maps, my AI takes the map design into account. To build a generic AI reacting to every map (e.g. islands) is almost impossible.
All maps: - Scouting was increased (4:00 to 4:30) - If a Phoenix dies a big Phoenix image is displayed - Gas is now build if an expansion is ready - Alterzim: Removed Ebay
I have played many, many games against the computer's Terran Elite A.I. and I actually think it has much to recommend it to get a beginning player to, say, gold level ladder play by making you build units efficiently and scout what units it is going to make.
The good things I would have to say about the Blizzard Elite Terran AI:
-It seems have fairly realistic and moderately aggressive timing for getting expansions and building workers, sometimes comparable to what good human players would do, this is improved from earlier HOTS versions which were very slow to take their second and third bases and would max out at around 50ish workers.
-If it does significant damage with an early attack it stays aggressive and actually can close out a game by continually applying pressure and not letting you back on your feet.
The bad:
-It does not make enough production facilities. So if you win a major engagement of maxed or near maxed armies, even if it has plenty of bases and economy it won't have enough barracks or factories to remax and you can win fairly easily just by remaxing a lot faster than the computer. Simply pumping out units should be something the computer could excel at. For a given unit composition (bio or mech), should it not have some algorithm in mind for how many factories or barracks/starports it should put down when it has a 2 base versus a 3, 4 and 5 base economy?
-It only maximizes research of bio upgrades, even when the build is clearly mostly mech units, Thors and hellbats, and it has built 4 factories and twin armories, it only researches +1 armor for its mech units. Even if bio units are only a small portion of its army it will research those upgrades to 3/3 over time, but it never seems to go beyond +1 armor for mech even when it has loads of money, unless the game goes very very long.
-I would think that an early 7 or 8 hellbat bum rush would be something the computer could execute fairly easily. The type that just beats down your front door as Zerg and doesn't rely on making medivacs or dropping. It should not require any micro really except transforming to hellbats right outside your base. It would actually be pretty valuable to me if the Terran Elite A.I. mixed in a build like this to practice trying to defend it with slow banelings vs. roaches vs. lots of queens and lings and spines. If that A.I. can do a cloaked banshee rush, and it occasionally does, it should certainly be able to do this.
EPT
![[image loading]](http://i.imgur.com/8Tu8EPd.jpg)
![[image loading]](http://i.imgur.com/1b2yLBK.jpg)
