|
On September 18 2015 06:58 [[Starlight]] wrote: The few times when a pro actually has to *guess* which start position to send the lings to, he always seems to guess right... as in this game with Luxury 4-pooling Mind. Is there anything Mind did to give away his position? I can't see anything. And Lux's OL doesn't have sight of Mind's main or nat in time.
It doesn't seem that strange to me. His drone scout eliminates one possibility and he sends the lings cross position because he would expect to see an earlier scv scout from the other base (although Mind happened to not scout that game at all). In other words, while he could've been wrong, it was a reasonable decision.
|
On September 23 2015 03:34 f10eqq wrote:Show nested quote +On September 18 2015 06:58 [[Starlight]] wrote: The few times when a pro actually has to *guess* which start position to send the lings to, he always seems to guess right... as in this game with Luxury 4-pooling Mind. Is there anything Mind did to give away his position? I can't see anything. And Lux's OL doesn't have sight of Mind's main or nat in time.
It doesn't seem that strange to me. His drone scout eliminates one possibility and he sends the lings cross position because he would expect to see an earlier scv scout from the other base (although Mind happened to not scout that game at all). In other words, while he could've been wrong, it was a reasonable decision. Oh yeah... standard enemy scout timing as a tell. I completely forgot about that.
So, noodling it out... from near positions, there's a 50% chance the 1st enemy scout would've gone the wrong way and Luxury wouldn't have seen any scout by the time of his rush (and his OL wasn't scouting along any enemy scout path, wouldn't have seen one).
But yes, it's an even higher chance of no-see from cross positions (which sorta makes me want to early cross-scout now just to screw ppl up ). Playing the odds, you send your rush cross-position.
(although... what about ppl who cross-position scout first to check center map for proxy gates/rax?)
Lux's very logical assumption would've still gotten him in trouble had Mind been at 1 o'clock (OL wasn't there yet at rush-time), but you're right, it makes perfect sense, and it worked out that time.
Damb, love all the subtleties of BW, the little things really matter... just a great design.
|
not a strategy question but you know..... around 2006-2008 there was a tournament between teams or nations, where each team had a bw planet (consisting of 5 maps of that tileset) and they could attack the other teams planet to gain control of it. anybody remember the name of the tournament, i cant seem to find it.
|
Since this has been happening too often for my liking, why do units sometimes refuse to move?
When I'm microing pretty hard and trying to kill a chasing unit for example, sometimes my units stop dead in their tracks and refuse to move for a few seconds.
Anyone knows what's going on or how I can avoid this?
I have a recent replay that can showcase if anyone wants to see it in action.
|
Kau
Canada3500 Posts
Sometimes they lock up if you try to move them mid attack (happens a lot with dragoons). Just press s to stop them and they should be able to move again.
|
It's because you're telling them to attack a new target right when it moves out of range. If they're firing at the same target as before, they will never freeze. This happens all the time with marines.
Edit: Oh, and the reason it happens is because there's a very short attack animation which happens the first time your unit selects a target, then is skipped in subsequent attacks. To avoid it you can either manually focus a specific unit to kill, or only a click on the ground when you're sure an enemy is in range. If it happens, hit S like Kau said. Just tap s as fast as you can, then issue the next command instantly.
|
I see it in bot games all the times. Marines that are going to an attack animation but never shoot because they switch targets mid-animation.
Use the techniques from the posts above me to solve this.
|
Is there a micro tournament map out there where you can pick a race and only have to micro units from that race?
And what is the best/most popular micro tournament map in general?
|
The most popular map is MicroTournament10.6h or .6n, and yes it lets you pick specific races. Pretty much everyone has the map, but I could send it to you if you want.
|
On October 11 2015 04:28 Sero wrote: The most popular map is MicroTournament10.6h or .6n, and yes it lets you pick specific races. Pretty much everyone has the map, but I could send it to you if you want.
That is exactly what I am looking for. I already got the map from http://sc.nibbits.com .
Didn't know that kovarex, the developer of the BWAPI, made BW maps as well.
|
On October 11 2015 08:06 LetaBot wrote:Show nested quote +On October 11 2015 04:28 Sero wrote: The most popular map is MicroTournament10.6h or .6n, and yes it lets you pick specific races. Pretty much everyone has the map, but I could send it to you if you want. That is exactly what I am looking for. I already got the map from http://sc.nibbits.com . Didn't know that kovarex, the developer of the BWAPI, made BW maps as well.
The map doesn't seem to have mirror match-ups though. When both select terran you get Terran vs Protoss matches. Is there a micro map that does have mirror matchups?
|
United States1431 Posts
Hey, some guy made a thread that said he had a bot running to record all of the broadcasts of a few select players(hero, effort, ssak, etc.) I can't seem to find that thread.
|
On October 12 2015 15:50 Ty2 wrote: Hey, some guy made a thread that said he had a bot running to record all of the broadcasts of a few select players(hero, effort, ssak, etc.) I can't seem to find that thread.
korhal.info.pl (i notice its not really working for EffOrt though)
|
In the micro tournament level 28, 9 marines (with range and stim) are put up against 12 cracklings. How does the terran win this engagement? Even with millisecond perfect hit and run micro (move when the weapon cooldown is in effect, shoot as soon as there is no cooldown) plenty of cracklings survive.
|
are the bot marines capable of precise focus firing, and zero overkill? (6 marine->insta kill 1 ling at every hit)
and if the zerg is doing proper micro (concave and surround before attacking) and not just simple amove, theres not a lot terran can do other than hit and run, avoiding surround for as long as possible
its micro tournament map where either side with better micro (less human errors) wins, not single player micro challenge map
|
United States1431 Posts
That stage has always been hard for me but I've seen some methods work better than others.
The first strategy is to peel away the layers of marines on the front and sides to the back while the other marines in the far back are firing. You have to do this in rapid succession much like how you'd begin to split your marines vs. psionic storm or an infested terran ball or something. If you know sc2, it's like micro vs. banelings which are like infested terrans. Of course, don't do it too fast or else marines won't be given the time to fire.
Also, you need to set the stage before the initial engagement. If you let the Zerg player get a position where he does a full picture perfect surround, you've lost. What you do is win the engagement before it even happens through good posturing. Once the zerglings do try to attack, stim, move the marines to the side so only one half of the zerglings are coming close to the marines. Through a representation of shapes it's like two rectangles' halves colliding. One half only manages to touch the other half where as the other halves of the rectangles hang over the edges on opposite ends. The marines are going to have the advantage due to being ranged units. That sounds pretty vague and I can't really express my thoughts that clearly. When I get home I'll whip up a microsoft paint image and upload it later.
Sometimes the zergling AI gets attracted to one of two halves of a marine group. This happens a lot actually and you can kind of bait this response by having one half very slightly protrude out more than the other half and with the previous method I stated above but sometimes the zerg kind of does it on its own and can vary if zerg kind of doesn't spread out zerglings evenly. You can move the other half that is being targeted by the zerglings while the other half picks off the zerglings. Sometimes you can run away the single marine that gets targeted by 3-4 lings so other marines can kill the zerglings as they chase that one marine. The zergling probably will redirect its target so then you can let the marine that was running around to start attacking again.
Going back to the micro of peeling away only the marines at the front and sides, it spreads out your marines in a semicircle formation that can help you better distinguish which marines are getting focused down and increasing the distance a zergling has to travel after killing off one marine. If marines are all bunched up, after killing one marine all the zergling needs to do is take one or two steps more and start chowing down on another marine as opposed to 5-6 steps if marines were spread out. Of course, you don't want your marines too spread out or zerglings will isolate marines without other marines being in range to fire and it'll be a unit lost in vain. This micro forces the zerg to micromanage more and zerglings might end up doing more moving than actual attacking. It only takes a few seconds and if Zerg doesn't react fast enough within 1 or 2 seconds it can easily mean a swift loss if they don't adjust properly or overcompensate.
You also want to stim before the zerglings are within firing range because time stimming only when zerg is close enough is time wasted because your units are going to be in normal speed/attack mode in that fraction of time and you'd rather have them stimmed for the very beginning as well to maximize efficiency and maneuverability/flexibility. I think that might be a problem for the AI.
Sometimes a human will move zerglings forward, act like they're engaging and force the stim out and then back off right away. This will force the terran to make a move before stim runs out in a frantic panic and it gives zerg the composure and time to choose when to engage while also running the possibility of having stim run out mid-fight. I guess that won't really be too big of a problem for AI.
The peeling away method is harder and usually the easiest way to win is to catch one half ot the zerglings off guard and abusing the AI targeting. You could try some combination.
|
nice writeup ty2 but honestly you are still mostly raise points from a human player perspective, im not a programmer (apart from some basic stuff) but imo AI do not have of a concept of "posturing" and "peeling" like humans intuitively do. they receive certain variables/stimuli (ling is approaching from E) and react based on what the algorithim churns out for them( initiate stutter step routine1 along W). what happens when human play with them, humans will start to notice that the AI follow certain predictable patterns, humans figure how to take advantage and then abuse it as much as possible for a easy win. letabot is obviously more advanced but there still a lot of room for improvement if it can't ever gain a win in that micro situation
their saving grace is that they are easily capable of some incredible mechanics/micro that humans cannot realistically do (ie. berkeley overmind). perhaps bot marines for instance can be tuned with smart targeting so that none of their shots are wasted unlike regular human control
honestly since we dont know much of the algorithim, letabot should post a vid of that scenario, it would be easier to see and dissect why the bot marines keep losing
|
My bot can win the fight against an a-move bot by simple camping in one of the corners. Not sure if this works against human players though.
This is an example of the bot fighting with hit&run (stutter-step in BW I guess) and always targeting the weakest zerling:
|
United States1431 Posts
Yo, that was good. It seems like the computer did some of the peeling action, retreating marines when they were getting targeted and attacking only when they weren't getting attacked. This level has always been tricky in my experience because zerglings win most of the time. I never did peel effectively. Can I see the prior 5 seconds? Did the marines just attack move parallel line to parallel line? I'm trying to think from a programmer perspective how you could make marines win. This algorithm would have to be applicable to a real game scenario too I'm assuming.
Can you have the marines move to an area that is going aside the area the zerglings are initially directed towards? Like in the gif, have the marines move up or down when zerg goes left . If zerg is going down have the marines move right or left so they're never going head to head. That might make it harder for zerg to evenly distribute zerglings. The marines would also have to react soon enough. Even if they can move only like three steps before lings get in firing range I think that might be a minor improvement.
The bot keeps letting the ling get good surface area . Second mode, have the marines pre emptively spread out semi circle and have them try the same kiting algorithm. Did you optimize the ling AI too or did you just use default SC AI? I think zergling distribution would be hard in this scenario too.
Third idea, have the marines be in a circle, just an actual circle of marines. Magic box or whatever so they retain their shape. Then, then, truly, the marines may win. Maybe experiment with different size circles and spacing apart of the marines? Somehow marines would have to detect soon enough in a real game that the zerglings are approaching in order to get in a circle-ish formation. Oval maybe?
My Fourth idea is to simply have the marines be a vertical line that goes perpendicular to the zerglings. Probably not going to work but worth a shot. I can log in maybe and help somehow or are you fine?
Fifth idea, have the marines assembled into two parallel lines. I think this could do the trick. Programming wise maybe you can tell the front row of marines to attack towards the zerglings and have the back marines stay still until they're in range of the zerglings? The kiting algorithm would take care of the rest.
|
This was just both marines and zerlings attack moving toward each other, with only the marines having some micro implemented in them.
I am going home right now and will be back online 23:00 CEST (2 hour 30 min from now). I can only host from university atm, so if you can host and help test the second and third option that would be great. There is also the option to keep the marines into a corner ( and sometime move them to another corner to prevent camping trigger).
|
|
|
|