Humanik V1.0.0

Started by AlsoKnownAs, April 09, 2010, 09:59:01 AM

Previous topic - Next topic

sur_real

Looking good, one problem though (not that big though), imo mineral fields are over saturated with probes (at one point in the video you can see ~30 probes on the unit count for natural only; who Humanik had no expo)

you can check this thread for more information http://www.teamliquid.net/forum/viewmessage.php?topic_id=113577

AlsoKnownAs

hi sur_real, thanks for commenting :D .

We have done our research on the saturation of harvesters in a town, basically each mineral can accommodate up to 3 gatherers. So an amount of 30 would be considered the point of saturation (8 minerals count and 2 vespene gas counts, 10 * 3 = 30), the extra workers are used for scouting (if no observers are build yet). Humanik is not entirely a script AI just like the others, it only expands after numerous conditions are met, a set of conditions has been coded by us, as you can see in the video, Strategy V0.31 does not saturate its amount of worker in a base like Humanik does, 2 bases of 30 workers would be equivalent to 1 + the 400 mineral extra to build on your army.

Also, Humanik mass its army base on the condition evaluated, in the video, it knows that the amount of army from its opponent diminishes after the "great" clash, hence it would be best that it mass its army (instead of expanding) to finish the game. Of course, the decision of massing an army is still arbitrary given various decisions that can be taken during the gameplay as programmed.

Artanis186

#32
FYI, you can click and hold a units portrait to lock the camera on it. Now back to watching the video. :)

EDIT: Most of my reply is on the second part of the video itself.

All I have to add beyond that is, one thing I noticed, the Sentries seemed to waste Gaurdian Shield a bit.

Next round, I request playing against someone else besides Zerg.

emylgoth

#33
No one finding this ridiculous? Is only talking for 4 days now. And your video comments are pretty childish and what is funny is that you almost always wrong type your own AI name "Humanik" but "Strategy" not. Why you're not recording your voice when you're making those videos? Maybe, in that way, you will inspire more reliability to all viewers (or not? :P ). If you want to have a real support from people here, you should post your first release with at least one race (protoss, I guess) and if it will be liked, people will search for your next releases and comment about it. My advice: don't push to much with theoretical explanations before you have something out.

P.S.
And... when you're recording, please set the game speed to fastest from the beginning...

klaretos

You are stupid emylgoth. You are implying that Humanik bot is not real? LOL! I am sure that when it comes out it will be the best non cheating bot around for a long time. The only one that will be comparable will be starcrack 8 and we don't even know when it will be ready. Humanik will be out very soon, we saw it in action.

One question AKA, you said that Protoss is the race to play against. How difficult Terran and Zerg are at the current state?

emylgoth

#35
Quote from: klaretos on April 13, 2010, 07:00:42 AM
You are stupid emylgoth. You are implying that Humanik bot is not real? LOL!

I said that?... Holly molly, I can't remember and I guess I need to work more harder on my English... Sorry if I upset someone.
All I tried to say is: release a first version of AI and then discuss it. Don't make such a long propaganda before people know it.
But... do your job, please, with Normal speed, since no one care about that.

itsarabbit

Quote from: AlsoKnownAs on April 12, 2010, 01:44:29 PM
itsarabbit, thanks for commenting :) .First of all, to correct some minor assumptions from you, a zealot is not able to clear 4 zerglings by itself + when there's 2 roaches firing at it, that would be more of an issue (And how would you know that the roaches ain't micro manage to avoid your focus fire? That would be a waste of time trying to shoot it while at the meantime having your zealot getting killed). Anyhow, the simulation of such a situation is not done by me but by my other team members and I'm only giving you an easy example compared to a more complex one.
AFAIK, if you would shoot at the roaches, 2 or 3 lings would at most be able to survive after the zealot would've gotten killed, and its easier to kill them at first, and if they are avoiding, they wouldn't be able to fire at the zealot.
and I said that the zealot would've almost clear 4 lings, only 1 would survive.
Anyways, there is no reason for them not to focusfire the immortal anyways, making this conversation useless.
Want Beta key! :D

AlsoKnownAs

#37
emylgoth,

QuoteNo one finding this ridiculous? Is only talking for 4 days now. And  your video comments are pretty childish and what is funny is that you  almost always wrong type your own AI name "Humanik" but "Strategy" not.  Why you're not recording your voice when you're making those videos?  Maybe, in that way, you will inspire more reliability to all viewers  (or not? :P  ). If you want to have a real support from people here, you should post  your first release with at least one race (protoss, I guess) and if it  will be liked, people will search for your next releases and comment  about it. My advice: don't push to much with theoretical explanations  before you have something out.

P.S.
And... when you're recording, please set the game speed to fastest from the beginning...

It's not my decision whether to release or not, I'm not the project leader, so get real. If you don't believe that it exist, go make one yourself which can possibly beat even the simplest AI out there and play with it yourself. If you don't like my comments, go to some other thread and post there. Plus, I'm not good at commenting with my voice, not everyone is born to be a good commentator. And how can you be sure that they are theoretical? Presuming that you're not one of the SC2 programmers, so don't assume something just because you "think so". Childish you say eh? I bet you can't even make one decent video, go get a life "kiddo", we are making this for free and not paid, so send your negative comments somewhere else. And about the speed of the video, I don't see anyone else complaining, so what's your problem?

itsarabbit,

As said in my previous post, the decision of having to prioritize (focus fire) the roaches or not is not under my jurisdiction, it is in progress (simulations and other set-ups), thanks for your comment anyways  :) .

P.S. I'm merely posting these videos so that you guys are having updates on our progress (similar to a developer's diary), cheers :cheers: . Also, we are programming at our own speed, no deadline, if you can't wait for the release, there's always other AIs to play around, if you don't bother whether it's going to be released, then don't post, simple as that.

gzxaaa

To the author of Humanik AI:

Please test your AI against Strategy AI Protoss & Terran 0.31 (NOT 0.13!). The zerg strategy AI is much weaker than protoss and terran (almost 100% lose if you do ZVP or ZVT). Protoss is strongest for Strategy AI 0.31.

Strategy is probably still the best at Macro (esp economy). Your AI expands too late, and 30 workers per base has almost no better income than 25 workers per base used in Strategy AI.

itsarabbit

Quote from: gzxaaa on April 13, 2010, 10:00:31 AM
30 workers per base has almost no better income than 25 workers per base used in Strategy AI.
Actually(Now I did not watch the video), 30 workers per base is ideal. 3 workers per mineral patch and vespen geyser, though it is better to split up your workers between expansions by sending atleast 8 as soon as you get one.
Want Beta key! :D

Chriamon

#40
Quote from: itsarabbit on April 13, 2010, 11:24:00 AM
Quote from: gzxaaa on April 13, 2010, 10:00:31 AM
30 workers per base has almost no better income than 25 workers per base used in Strategy AI.
Actually(Now I did not watch the video), 30 workers per base is ideal. 3 workers per mineral patch and vespen geyser, though it is better to split up your workers between expansions by sending atleast 8 as soon as you get one.
Actually, You are kind of wrong. Passed 16(2 per mineral patch), there is hardly any benefit to increasing workers, and past 24 (3 per patch) there is absolutely 0 benefit to increading. At 16 workers, you get something like 17 mins per second, and at 24, its like 18 or 19 mins/sec. Theres a whole thread about it on TL. The reason you keep building past 16, is if you are planning on making an expansion, so that when the expo is done, you have those 8 workers to send. (Note, you will have +3 more workers per gas you are using, which is why a lot of times there are ~22 workers per base). 30 per base is the max benefit, but more like 22 is ideal.

Also, keep posting AKA, it is interesting to know your progress, although, like people have said, it would be nice to see more examples of it in action.

And, about whoever was talking about strategy .31 zerg AI, it is not bad, it only loses to the protoss and terran start AI because the terran and protoss both go for all-in strats. They are both extremely simple to beat if you hold off their initial attack. If the strat .31 zerg is not quickly rushed (relatively quickly), it can build up a sizable force and out macro the opponent quite easily.

Hadrian

#41
As far as workers are concerned, the ideal number is 22-24.

As a general rule, you build 2 workers for each crystal and 2-3 (based on distance) for every vespin gysir.
Any more will just clutter your ressource field and drain ressources, and as a Terran, you should always have 1-2 workers less so that there is space for your mules.

Every worker needs time to be build and costs 50 minerals. If you produce for example 35 instead of 21 workers (terran base, 1 vespin gysir only needs 2 for optimum), you have wasted 14x50=700 minerals for little to no benefit.
Yes, someday you might need them in your next expansion, but until then, those minerals would have been useful to upgrade your troops, or instead, build 14 marines, or 28 zerglings, or 5 marauders, or a few supply depots and bunkers, etc.
Oh, and - every worker needs supply. This in time forces you to buy (waste more minerals) on supply buildings/units you otherwise would not have to. And since there is a unit maximum, it's a good idea to keep the workers balanced and to generally have as few as you need and nothing above neccessary.

The idea is not to put as many workers as possible for a bit more efficiancy into your base, but to give every single worker a purpose for existance and to utilize every crystal you have in order to push an early techno+unit advantage. That is why it's not a good idea two build past 22/24. It will take minutes for those workers to actually get the crystals back that you used to build them.
Thus, why not leave them be and use the ressources instead for much better purposes?

B_w_S

Quote from: Hadrian on April 14, 2010, 06:30:54 AM
As far as workers are concerned, the ideal number is 22-24.

As a general rule, you build 2 workers for each crystal and 2-3 (based on distance) for every vespin gysir.
Any more will just clutter your ressource field and drain ressources, and as a Terran, you should always have 1-2 workers less so that there is space for your mules.

Mostly right. . . . .

Pro tip: Mules don't hog any space from scvs they can both mine the same patch at the same time.

Therefore 22 (16 on minerals + 6 on vespene) is the most efficient amount for mining unless you plan on Maynarding any time soon.

AlsoKnownAs

A message from one of our project organizers;

QuoteDamn, the engine is way too medieval!

Facts:

QuoteThe engine provided by SC2 Beta for AI integration is limited  and not flexible (yet). (Example: There is no function provided to us that  enables us to do a custom placement of units, especially buildings  based on the desired points) And since its a script, there's nothing much that we can do about it.

It is crucial for us to know that advanced players will not just micro manage its army and economy but also take advantage of the terrain provided. The engine that is implemented within SC2 Beta has not emphasized terrain in mind. (The path blocking that you watch mostly in replays are not implementable because there is no function in the current engine which enables us to place buildings at our own discretion but instead based on pre-defined build flags) - and these predefined functions are a pain in the arse (e.g. it seems like for each pylon, the maximum building that it can accommodate is between 14 to 15, based on the original SC2 programmers)

Other strategies that you watch in replays are implementable except the ones that utilizes on the placement of buildings (Path blocking, dummy, seperation of production facilities, bunker rush, photon cannon rush, and etc.).

The above is one of the minor drawbacks that we are encountering  :( , the current AI will never able to surpass a human if such limitation exist. Dispite so, we are working on implementing other strategies first and put those that are not implementable aside and have them coded once the engine improves.

whoamI

are you sure about buildings placement? if you watch strategy AI terran replays you will find out that the AI is blocking command center with buildings and uses supply depots as gates.. something that is pretty smart..