Zerg Build Order optimizer. - Page 26

azzu

Germany141 Posts

October 23 2010 00:15 GMT

#501

On October 23 2010 02:32 Cirqueenflex wrote:
...
it took my computer about 5 minutes to figure out the best build
...
after 5 minutes i got the answer that it takes 3 seconds more if i want to get a queen in this build,

5 minutes is not even remotely near the best build. You can be about 99% sure its not the best build. Maybe if you run 200 million games you can be 80% sure that you found the best build. (numbers just for the sake of explaining, no warranty)

On October 23 2010 02:32 Cirqueenflex wrote:
Until 25kk games it just calculated games where it never even "thought" of the upgrade, went through the builds and then suddenly wondered why it didn't get the required upgrade in time etc...
The next 5kk games brought out 1 build to meet the requirement. Until my work is done i will let it calculate the result but i feel like this takes days

this propably came because you wanted 1 queen before 3:00 minutes, which requires a 6 or 7pool, and doing that is so rare that it has to compute many many builds before it even gets to the point of getting an upgrade.

Ghostlike

United States4 Posts

October 23 2010 00:33 GMT

#502

Wow, take all skill from the game. This is nothing more than a modern cheat code. IMO this garbage shouldn't even exist.

Edit: This is the equivalent of spreadsheets for WoW. They do nothing but groom bad players.

dotFX

United States131 Posts

October 23 2010 00:40 GMT

#503

On October 23 2010 09:33 Ghostlike wrote:
Wow, take all skill from the game. This is nothing more than a modern cheat code. IMO this garbage shouldn't even exist.

Edit: This is the equivalent of spreadsheets for WoW. They do nothing but groom bad players.

so doing addition and subtraction is equivalent to a cheat code? or just when a calculator does it?

im confused bro....

tetracycloide

295 Posts

October 23 2010 00:59 GMT

#504

Builds, much like perfectly optimized gear, do not take 'skill' to obtain. They take time. There's no reason to spend hours of your own time in a build order simulator to figure something out like a machine when a machine can do it for you. Once you have a perfectly optimized build you still have to execute it which requires skill, focus, and no small amount of practice. Beyond that you must be able to adapt it to the situation at hand, which requires skill, intuition, and perception. Really this doesn't remove any aspect of the game that's 'skill' based at all.

What's wrong with grooming bad players? Why wouldn't improving the play of a 'bad' player by removing obvious flaws be a good thing? I mean that's what groom means, cleaning up, so what's bad wrong with that?

chumpchous

68 Posts

October 23 2010 01:04 GMT

#505

There seems to be a bug where it is overrewarded for producing overlords. After enough time, it eventually corrects the problem, but I feel like you could reduce a significant amount of processing time if it didnt build up to 352 supply cap over and over.

bovi

Japan208 Posts

October 23 2010 01:17 GMT

#506

so i guess copying any new build out there or strat you see on GSL is a cheat too? get off your self made throne.

--------------------------------

I'm liking this and support it. now to learn how to use it

Neo.NEt

United States785 Posts

October 23 2010 02:45 GMT

#507

Have you guys tried to do these builds? I just tried a 10 roach in 5:40 build with some drone constraints and the fastest I was able to do it was about 6:10. Granted I only tried twice, but I didn't make 30 seconds worth of mistakes. I'm assuming it doesn't count drone moving time? I was about 15 gas short of being perfect both times, either from user error or because in the program the drones can teleport to an extractor. Should I assume that it actually takes a little longer to do these builds than it says or was I just bad? Still at 6:10, this was a hell of a lot better build than I was ever doing...

tetracycloide

295 Posts

October 23 2010 02:51 GMT

#508

So far I can think of three enhancements that would greatly improve the end user experience:

1. Rest button. It would be nice if there was a button that simply cleared out all the settings and stored builds so the program could start over from scratch without actually having to be closed and restarted

2. Maximums as well as minimums. Currently the user can only say 'build at least x of unit/buiding/upgrade y' and it would be a huge improvement in both usefulness and possibly processing speed for the user to say 'build no more than x of unit/building/upgrade y.'

3. Resource counts as inputs. This might just be something I want that's not useful for everyone but I'd really like the option to add a specific amount of banked resources as a build requirement. This will improve the end-user experience by more explicitly giving them what they want. A 5 roach rush is, more accurately, a roach warren, 375 minerals, 125 gas, and 5 larva. It's important in a rush style build to hide or simply not build tech until the last possible second and to only reveal it once the units arrive. If the calculator searches for a 5 roach rush that pops a warren then builds 3 roaches, then 1 more, then 1 more it's not as good even if it gets all five out slightly faster because the window were the rush can be scouted is bigger.

azzu

Germany141 Posts

October 23 2010 03:31 GMT

#509

On October 23 2010 11:51 tetracycloide wrote:
If the calculator searches for a 5 roach rush that pops a warren then builds 3 roaches, then 1 more, then 1 more it's not as good even if it gets all five out slightly faster because the window were the rush can be scouted is bigger.

Actually every rush i calculated with this program gets the units at the very very last second all at once.

Perscienter

957 Posts

October 23 2010 04:05 GMT

#510

I have the impression that this tool just applies the algorithm in a primitive way, so the algorithm needs fine-tuning.

After 6k games it is saying that 9 hatch would be faster than 6 hatch. That's not correct.

cup of joe

28 Posts

October 23 2010 04:06 GMT

#511

On October 23 2010 05:20 icezar wrote:
Another Perf tweak could be Q, there is no point in more that 1q/base unless it is a requirement. Currently it treys things like 2 base 4q ?

there actually is a reason for multiple queens per hatch, with absolutely perfect macro the larva pop when you have 23-24 energy on the queen so adding another could save like a second each inject

dunno if thats why it does it or not but technically it is better

tetracycloide

295 Posts

October 23 2010 04:07 GMT

#512

On October 23 2010 12:31 azzu wrote:

Show nested quote +

Actually every rush i calculated with this program gets the units at the very very last second all at once.

How many rushes have you tried? All of them? That's really only one use of the feature anyway. It could also be used to create reactive builds. A resource target at time x would allow the resources to be used on anything. As an example I might know I want roaches and lings vs. a 4 gate at around 6:00-6:30 but I won't know for sure how many of each until I see the zelot/stalker/sentry count I'm facing so it's better to have 8 larva, some amount of resources, 16 free supply, a roach warren, and a spawning pool at ~6:00 then it is to say 8 roaches at 6:30 or 16 lings at 6:30

Dee-Kej

Sweden191 Posts

October 23 2010 05:33 GMT

#513

So I've been going at it for just maybe ~1h, testing and letting the program do its thing, and there's a few things I'm noticing that could be changed for the better.

1. The build is not done, but it is showing me that I need 15 hatcheries. 15. 5 of them at the end of the build. I don't know about you guys, but I can't seem to recall any 1v1 map that allows 15 expansions.

Being able to put a maximum number of things would be good.

2. Priorities. The way the program test builds is far from optimal (duh, it's an early version). It fixed that right now but, 5 minutes ago, it created 10 of the 40 final drones in the last couple of minutes. Plus, before I put in a requirement for 40 drones, it only created 12 drones, and tried to execute a 30 minute build with that. It doesn't take a genius to figure out that you need more than 12 drones for a build to be optimal in the long run. So some sort of function that allows you to "help it along" would be sweet, and priorities could be an option?

3. Supply cap. Sometimes it goes over 200/200, which of course is acceptable if units dies, but having ~30 overlords for XXX/300 is a bit much, no?

4. Something that allows you to set how much % of the cpu the program is allowed to use. Right now it runs at up to 95%, which isn't exactly great for the comp afaik.

Otherwise it seems like a damn good program, and as far as I can tell it's a good UI for such an early build of the program. And I love how it keeps going along the path it ended at even after starting after you have stopped it to change something small in the waypoints.

Edit: Just to add - the build I'm testing isn't done by far, so my problems might disappear in time, but these are just things I've noticed that could make the process a bit faster. It's such an awesome idea, and I wanna help in any way I can.

Edit2: The build I tested was a 30 min build with 50 ling, 5 ultra, 20 muta, roaches and full upgrades (random numbers). Testing right now instead the fastest way for 30 speedlings and 7 speedroach with expansion, and is going satisfied at under 8 min.

Fodder03

Canada142 Posts

October 23 2010 07:39 GMT

#514

Im trying to remake the 3 hatch muta/ling from sc1 (was my fave and VERY effective build). Always have problems as to when i want the 3rd hatch to go down in sc2 however, hence why im using this program.

Edit: Added a 6 minute waypoint (rough timing on 4 gates as far as i know)
23 drones
2 hatches
2 queens
10 speed lings
2 spines
It keeps failing on this waypoint, does this seem like too much to ask for the 6 minute mark?

I have it set up to have:

limit of 10 min
46 drones (full saturation, is this overkill?)
2 hatches, 1 lair
1 evo chamber
3 queens
4 spines
12 lings with speed
and 4 extractors.

It seems to only want to build 1 extractor, and spines going down around the 7 minute mark, clearly not in time for a 4 gate or a stim timing push.
Any ideas or suggestions as to how i can make this work better? manipulate the timings on certain things that NEED to happen faster?

The idea behind the 3 hat muta/ling was to use sunkens and lings to defend early pressure and then mass up mutas to harass and take a 3rd. If u got an advantage u could pump muta/ling non stop from 2 base (3 hatch) all in/heavy pressure style and win the game. I would like to have the same sort of build for sc2 as my standard opener vs t/p.

any help please?

(I think it is getting hung up on number of extractors to be honest, possible bug? 90/298 supply? ... really? lol )

icezar

Germany240 Posts

October 23 2010 07:52 GMT

#515

On October 23 2010 08:55 azzu wrote:

Show nested quote +

post the exact BO please

I do not have it anymore as i changed the ver. from 0.14 to 0.15 and forgot to save.... STUPID!
Anyway i put it again in 0.15 and it got even better

44drones 1Hatch 1Lair 2Q 1Extractor -> 5:57! and it starts with an overpool?!
at the end i have 445min and 186gas - i think it can squueze in speed RW HD and some spines without affecting the time too much, i will try that later.

+ Show Spoiler +

@0:00 M:50 G:0 L:3 S:6/10 BuildDrone
@0:12 M:51 G:0 L:2 S:7/10 BuildDrone
@0:15 Spawned: Larva+1
@0:17 Spawned: Drone+1
@0:23 M:53 G:0 L:2 S:8/10 BuildDrone
@0:29 Spawned: Drone+1
@0:30 Spawned: Larva+1
@0:32 M:50 G:0 L:2 S

/10 BuildDrone
@0:40 Spawned: Drone+1
@0:45 Spawned: Larva+1
@0:49 Spawned: Drone+1
@0:49 M:104 G:0 L:2 S:10/10 BuildOverlord
@1:00 Spawned: Larva+1
@1:14 Spawned: Overlord+1
@1:15 Spawned: Larva+1
@1:17 M:202 G:0 L:3 S:10/18 BuildSpawningPool
@1:25 M:53 G:0 L:3 S

/18 BuildDrone
@1:33 M:54 G:0 L:2 S:10/18 BuildDrone
@1:40 Spawned: Larva+1
@1:41 M:55 G:0 L:2 S:11/18 BuildDrone
@1:42 Spawned: Drone+1
@1:48 M:54 G:0 L:1 S:12/18 BuildDrone
@1:50 Spawned: Drone+1
@1:55 Spawned: Larva+1
@1:55 M:59 G:0 L:1 S:13/18 BuildDrone
@1:58 Spawned: Drone+1
@2:05 Spawned: Drone+1
@2:10 Spawned: Larva+1
@2:10 M:140 G:0 L:1 S:14/18 BuildDrone
@2:12 Spawned: Drone+1
@2:22 Spawned: Spawning Pool+1
@2:25 Spawned: Larva+1
@2:25 M:238 G:0 L:1 S:15/18 BuildDrone
@2:25 M:188 G:0 L:0 S:16/18 BuildQueen
@2:27 Spawned: Drone+1
@2:40 Spawned: Larva+1
@2:42 Spawned: Drone+1
@2:50 M:307 G:0 L:1 S:18/18 BuildHatchery
@2:55 Spawned: Larva+1
@2:55 M:60 G:0 L:2 S:17/18 BuildDrone
@2:57 M:31 G:0 L:1 S:18/18 BuildExtractor
@3:02 M:55 G:0 L:1 S:17/18 BuildDrone
@3:10 Spawned: Larva+1
@3:12 Spawned: Drone+1
@3:12 M:105 G:0 L:1 S:18/18 BuildOverlord
@3:15 Spawned: Queen+1
@3:19 Spawned: Drone+1
@3:25 Spawned: Larva+1
@3:27 Spawned: Extractor+1
@3:37 Spawned: Overlord+1
@3:37 M:260 G:20 L:1 S:18/26 BuildDrone
@3:40 Spawned: Larva+1
@3:40 M:237 G:26 L:1 S:19/26 BuildDrone
@3:40 M:187 G:26 L:0 S:20/26 BuildQueen
@3:54 Spawned: Drone+1
@3:55 Spawned: Larva+1
@3:55 M:177 G:55 L:1 S:22/26 BuildOverlord
@3:57 Spawned: Drone+1
@4:00 Spawned: Larva+4
@4:00 M:129 G:64 L:4 S:22/26 BuildDrone
@4:00 M:79 G:64 L:3 S:23/26 BuildDrone
@4:02 M:51 G:68 L:2 S:24/26 BuildDrone
@4:07 M:53 G:77 L:1 S:25/26 BuildDrone
@4:10 Spawned: Larva+1
@4:17 Spawned: Drone+1
@4:17 Spawned: Drone+1
@4:19 Spawned: Drone+1
@4:20 Spawned: Overlord+1
@4:20 M:144 G:102 L:1 S:26/34 BuildDrone
@4:24 Spawned: Drone+1
@4:25 Spawned: Larva+1
@4:25 M:153 G:112 L:1 S:27/34 BuildDrone
@4:30 Spawned: Hatchery+1
@4:30 Spawned: Queen+1
@4:30 M:163 G:121 L:0 S:28/36 BuildLair
@4:37 Spawned: Drone+1
@4:40 Spawned: Larva+1
@4:40 M:151 G:40 L:2 S:28/36 BuildDrone
@4:40 M:101 G:40 L:1 S:29/36 BuildDrone
@4:42 Spawned: Drone+1
@4:45 Spawned: Larva+4
@4:45 M:125 G:50 L:4 S:30/36 BuildDrone
@4:45 M:75 G:50 L:3 S:31/36 BuildDrone
@4:50 M:100 G:59 L:2 S:32/36 BuildOverlord
@4:54 M:59 G:67 L:1 S:32/36 BuildDrone
@4:55 Spawned: Larva+1
@4:57 Spawned: Drone+1
@4:57 Spawned: Drone+1
@4:57 M:56 G:72 L:2 S:33/36 BuildDrone
@5:00 M:55 G:78 L:1 S:34/36 BuildDrone
@5:02 Spawned: Drone+1
@5:02 Spawned: Drone+1
@5:10 Spawned: Larva+1
@5:10 M:182 G

7 L:2 S:35/36 BuildOverlord
@5:10 M:82 G

7 L:1 S:35/36 BuildDrone
@5:11 Spawned: Drone+1
@5:14 Spawned: Drone+1
@5:15 Spawned: Larva+4
@5:15 Spawned: Overlord+1
@5:15 M:126 G:107 L:4 S:36/44 BuildDrone
@5:15 M:76 G:107 L:3 S:37/44 BuildDrone
@5:17 Spawned: Drone+1
@5:17 M:65 G:110 L:2 S:38/44 BuildDrone
@5:19 M:54 G:114 L:1 S:39/44 BuildDrone
@5:25 Spawned: Larva+1
@5:25 M:123 G:126 L:2 S:40/44 BuildDrone
@5:25 M:73 G:126 L:1 S:41/44 BuildDrone
@5:27 Spawned: Drone+1
@5:30 Spawned: Larva+4
@5:30 M:125 G:135 L:4 S:42/44 BuildDrone
@5:30 M:75 G:135 L:3 S:43/44 BuildDrone
@5:32 Spawned: Drone+1
@5:32 Spawned: Drone+1
@5:34 Spawned: Drone+1
@5:35 Spawned: Overlord+1
@5:35 M:134 G:145 L:2 S:44/52 BuildDrone
@5:35 M:84 G:145 L:1 S:45/52 BuildDrone
@5:36 Spawned: Drone+1
@5:40 Spawned: Larva+1
@5:40 M:147 G:154 L:2 S:46/52 BuildDrone
@5:40 M

7 G:154 L:1 S:47/52 BuildDrone
@5:42 Spawned: Drone+1
@5:42 Spawned: Drone+1
@5:47 Spawned: Drone+1
@5:47 Spawned: Drone+1
@5:50 Spawned: Lairs+1
@5:52 Spawned: Drone+1
@5:52 Spawned: Drone+1
@5:55 Spawned: Larva+1
@5:57 Spawned: Drone+1
@5:57 Spawned: Drone+1
Satisfied.
Number of actions in build order: 60
---Final Output---
At time: 5:57
Minerals: 445 Gas: 186 Supply: 48/52
Drones: 44
Overlords: 6
Queens: 2
Hatcheries: 1
Lairs: 1
Gas Extractors: 1
Spawning Pools: 1
------------------
Sat Oct 23 09:40:09 CEST 2010: 3153.999525810324

But i have 41 drones on mineral with 2 base this means more than 16/base, i think the program does not know how to calculate the mining time lost.

Edit: just sow that the build ends with building last 8 drone 15 sec and it is caped by larva and not minerals so the build stands, i will try it in yabot to see how it goes, usually i am off by 5 sec every min so i should be able to do this before 6:25 no problem!

Edit2: I try it in Yabot, i am able to do this by 6:27 - last drone pops at this time, but the build does not seem optimal?!?
+ Show Spoiler +

100 Econ Build | 11 | icezar | BO to get ASAP 44drones 1Hatch 1Layer 1Extractor 10 100 0 56 1 1 31 0 | 10 210 0 87 1 0 41 0 | 15 180 0 153 1 1 32 0 | 18 325 0 190 1 0 36 0 | 18 60 0 199 1 0 35 0 | 18 100 0 210 1 1 31 0 | 20 165 8 238 1 1 32 0 | 22 100 28 247 1 1 31 0 | 28 15 16 294 1 2 3 0 | 32 115 64 318 1 1 31 0 | 35 100 100 336 1 1 31 0

Cambam

United States360 Posts

October 23 2010 08:54 GMT

#516

I don't know if this has been mentioned yet, but how hard would it be to optimize for unit number instead of build order time?

For instance, at the start you would choose the parameters: 10 minutes and roaches and then it would find the build order that produces the most roaches by the 10 minute mark.

I think this would be extremely useful for figuring out good timing pushes as well as figuring out good timing defenses. For instance, let's say you know a 15 FE nexus protoss is weak against mutas until 9:00; you would simply enter in "8:30" and "mutalisk" and the program would spit out the build for the most possible mutalisks by that time. Same thing with 4 gate defense. Find out the timing of the 4 gate push, then optimize for lings and roaches or lings and hydras until then.

Hopefully it's easy to code. If not, I still think it should be considered for something down the road.

Pwere

Canada1556 Posts

October 23 2010 09:22 GMT

#517

I agree with Cambam, and it should be easy to code, but the speed will suffer as it will get stuck in local maximas a lot.

redviper

Pakistan2333 Posts

October 23 2010 10:55 GMT

#518

On October 23 2010 14:33 Dee-Kej wrote:
So I've been going at it for just maybe ~1h, testing and letting the program do its thing, and there's a few things I'm noticing that could be changed for the better.

1. The build is not done, but it is showing me that I need 15 hatcheries. 15. 5 of them at the end of the build. I don't know about you guys, but I can't seem to recall any 1v1 map that allows 15 expansions.

That is a feature of genetic algorithms. The way genetic evolution works is that you mutate and assign a utility function. If this algorithm was doing a full search it would be far too costly (the build optimizer is funnily enough a TSP and so unsolvable for any problem of reasonable complexity).
By evolving the position in the solution space while at the same time using mutations to get out of local maxima you will be able to find a reasonably optimal build. I should point out that there is no guarantee that it will find a perfectly optimal solution - but it will find a satifiable solution eventually if one exists. (For values of eventually approaching infinity).

Now I haven't looked at the code myself but I do know a bit about these kinds of algorithms. Leavef it running nad it will eventually satisfy the conditions you give it. Either that or it will approach satisfiability.

Now one issue that I found is that it doesn't detect minimum bounds. It should immediately mark that WP as invalid if the request is impossible. That is, it should immediately calculate the theoretical minimum bound for the requirements.

redviper

Pakistan2333 Posts

October 23 2010 11:00 GMT

#519

Also Lomilar great work. The program works really well, has good core utilization and even the interface is nice. Are you a student or a professional developer?

Dionyseus

United States2068 Posts

October 23 2010 13:01 GMT

#520

On October 23 2010 20:00 redviper wrote:
Also Lomilar great work. The program works really well, has good core utilization and even the interface is nice. Are you a student or a professional developer?

I'm betting he's a pro. The program reminds me of a chess engine, and the fact that it supports multiprocessors is impressive, I know some chess engine authors who haven't been able to get their engine to utilize multiprocessor support.

Prev 1 24 25 26 27 28 58 Next

Please or register to reply.

Zerg Build Order optimizer. - Page 26

Completed

Ongoing

Upcoming