|
On October 27 2010 00:45 HanSoo wrote:Show nested quote +On October 26 2010 14:36 Bumblebees wrote:
It turns out that after some testing it fails pretty hard for more than 1 hatchery.
Only taking into account the production that a single hatchery would produce means that when you add a second hatchery, it will not maintain worker parity with a terran who has 2 command centers.
There is another better solution being discussed that's a combination of the worker parity idea and drone weighting.
btw, if I was better with the ide's svn plugin I'm using, I'd have explicitly noted your contribution to the idea. Yeah, I kind of realized that. Once the problem is expanded to more than one hatchery the problem gets much more complex. (Notice how in a real game when a zerg FEs their drone count is very low for a little while until the effects of the greater economy is able to kick in)
I just figured out a really clever way of fixing this issue.
I'm still running tests, but it seems to both solve the problem and encourage better hatchery timing.
|
On October 27 2010 00:48 Huumy wrote:Hello. I downloaded the version 0017 of this program and I have two question: Where do I find the disable all-ins button? I seriously can't find this but I'm sure it's obvious where it was hiding once I do. Is there place to post some of the cool BOs you find with this? Thanks for future replies 
It was a ghetto fix I added myself (I'm not one of the developers working on the project).
http://www.teamliquid.net/forum/viewpost.php?post_id=6726632
It's still available for download there, and in version 0018 (not released) you'll probably see an even better implementation of this as bumblebees is working on it now.
|
Holy crap this program is awesome. It even gets a rudimentary build order in just a few minutes. Hell you can just take that and adjust it as you see fit.
I notice in a lot of the earlier build orders the computer seems to favor getting overlords really quickly. I don't know if there's a way to adjust that, but I thought I'd mention it. I'm seeing a lot of times where it suggests an overlord at 13/18 and 29/34 etc. which seems rather early to me. I dunno, it's a possibility in terms of making it go faster.
Edit: The program doesn't know that you can only spawn one nydus worm at a time.
|
mad props to all you guys for sticking to the project and implementing serious features and bug fixes
|
Finland126 Posts
I notice in a lot of the earlier build orders the computer seems to favor getting overlords really quickly. I don't know if there's a way to adjust that, but I thought I'd mention it. I'm seeing a lot of times where it suggests an overlord at 13/18 and 29/34 etc. which seems rather early to me. I dunno, it's a possibility in terms of making it go faster.
You just need to give it more time and it will optimize your build better, I had my baneling, nudys, spinecrawler thing running over 5 hours and it still found ways to get it faster.
|
On October 27 2010 03:32 Huumy wrote:Show nested quote +I notice in a lot of the earlier build orders the computer seems to favor getting overlords really quickly. I don't know if there's a way to adjust that, but I thought I'd mention it. I'm seeing a lot of times where it suggests an overlord at 13/18 and 29/34 etc. which seems rather early to me. I dunno, it's a possibility in terms of making it go faster. You just need to give it more time and it will optimize your build better, I had my baneling, nudys, spinecrawler thing running over 5 hours and it still found ways to get it faster.
Yes I understand. I was just talking about ways the program could be theoretically sped up. I have no idea if putting more blocks and suggestions into it whether or not it would slow it down or speed it up or what. I would totally understand if that would be infeasible considering how awesome this program already is.
I also noticed that a lot of the games that it generates have multiple of the same tech structure. Obviously zerg doesn't need this unless they're going for multiple upgrades.
I just got an insanely wicked nydus rush timing from this optimizer. I'm going to try to test it tonight.
|
13 roach rush
131 million games played
last update: 9 hours ago
10 Overlord M:106 G:0
10 ExtractorTrick M:76 G:0
15 SpawningPool M:204 G:0
15 Extractor M:56 G:0
16 +1 Drone on gas M:214 G:0
16 +1 Drone on gas M:214 G:0
16 +1 Drone on gas M:214 G:0
18 RoachWarren M:378 G:55
17 ExtractorTrick M:228 G:55
16 Queen M:203 G:55
19 Overlord M:208 G:83
19 Overlord M:254 G:112
19 Overlord M:303 G:140
19 Overlord M:351 G:169
19 Roach M:400 G:197
21 Roach M:473 G:201
23 Roach M:448 G:185
25 Roach M:373 G:160
27 Roach M:298 G:135
29 Roach M:223 G:110
31 Roach M:247 G:104
33 Extractor M:172 G:79
32 Roach M:286 G:108
34 Roach M:349 G:111
36 +1 Drone on gas M:274 G:86
36 Roach M:317 G:100
38 Roach M:242 G:75
40 Roach M:167 G:50
42 Roach M 2 G:25
---Final Output---
At time: 5:47
Minerals: 200 Gas: 75 Supply: 45/50
Drones: 17
Overlords: 6
Queens: 1
Roaches: 13
Hatcheries: 1
Gas Extractors: 2
Spawning Pools: 1
Roach Warrens: 1
------------------
|
So the build I like the program to try the most is:
4 lings
17 drones
2 queens
2 hatches
0 everything else
no extractor trick
After 8 million games it starting building another ling (2) without being told to. And using that setup when I use the minimum build time of either pool or hatch or both at 14 it will be like 6 overlords needlessly.
Hope that helps you guys reproduce what I am getting.
edit: 15mil games later it stopped building 6 lings. Still seems odd that it would even try given my constraint.
edit2: at 20mil games it's now building an extra overlord.
|
On October 27 2010 04:39 Dudemeister wrote:
13 roach rush
seams very problematic for any fast teching T/P to deal with. gj just have to hide it long enough
|
On October 27 2010 05:15 teemh wrote:
So the build I like the program to try the most is:
4 lings
17 drones
2 queens
2 hatches
0 everything else
no extractor trick
After 8 million games it starting building another ling (2) without being told to. And using that setup when I use the minimum build time of either pool or hatch or both at 14 it will be like 6 overlords needlessly.
Hope that helps you guys reproduce what I am getting.
edit: 15mil games later it stopped building 6 lings. Still seems odd that it would even try given my constraint.
edit2: at 20mil games it's now building an extra overlord.
i guess your build has plenty of larva and minerals because 17 drones + 4 zerglings are built very fast, but 2 hatcheries and 2 queens take a loooong time to build.
so it just uses those larva and minerals to do something useful, instead of stockpiling the money.
with so few requirements you will propably already have the best build, it just is figuring out what extra things it could build in that time, like 2 zerglings or an overlord.
isn't the program clever? :D
|
Does it count larva for multiple hatcheries? It doesn't seem to...
+ Show Spoiler +
@3:54 Spawned: Hatchery+1
@3:54 Spawned: Larva+1
@3:54 M:541 G:0 L:2 S:19/28 BuildDrone
@3:54 M:491 G:0 L:1 S:20/28 BuildDrone
@3:54 M:441 G:0 L:0 S:21/28 BuildQueen
@3:56 Spawned: Drone+1
@4:09 Spawned: Larva+1
@4:09 M:491 G:0 L:2 S:23/28 BuildDrone
@4:09 M:441 G:0 L:1 S:24/28 BuildOverlord
@4:11 Spawned: Drone+1
@4:11 Spawned: Drone+1
@4:24 Spawned: Larva+1
Larva comes out every 15 seconds, even with 2 hatches. (I just told it to get me 5 hatches and 100 drones)
|
I don't know if you'll ever read this, hopefully you do.
I am just curious, what sort of learner did you use for the program? I'm trying to find any sort of documentation for the algorithm but had no luck =/
Why do i want to know? I am just so very curious (I'm a machine learningNLP research student).
|
On October 15 2010 04:10 puzzl wrote:
With the graph based approach and decent pruning you should be able to "solve" zerg up till at least 10 minutes without too much in the way of difficulty, and in doing so the program would only need to run a single time and could output any and all pertinent scenarios along the way, not just a single end goal.
My thoughts exactly. I don't see how this is a genetics problem. It seems like a graph theory problem.
I also realize that is just for fun and your hobby project is by above all the main reason you are doing this and using this particular method.
EDIT: Just thought about this for a while. If modeled correctly, I'm pretty sure this can be solved with a minmax search.
|
On October 27 2010 17:53 jdelator wrote:Show nested quote +On October 15 2010 04:10 puzzl wrote:
With the graph based approach and decent pruning you should be able to "solve" zerg up till at least 10 minutes without too much in the way of difficulty, and in doing so the program would only need to run a single time and could output any and all pertinent scenarios along the way, not just a single end goal. My thoughts exactly. I don't see how this is a genetics problem. It seems like a graph theory problem. I also realize that is just for fun and your hobby project is by above all the main reason you are doing this and using this particular method. EDIT: Just thought about this for a while. If modeled correctly, I'm pretty sure this can be solved with a minmax search.
You're just throwing keywords around that you heard somewhere. Minmax is nowhere near suitable for this. Minmax evaluates your gain vs your opponents gain for each choice. Graphs have loops in them which is not what you want here, you want a tree.
You want to use either GA (like the OP has) or you want a tree with A* heuristic expansion.
|
|
|
+ Show Spoiler +10 Overlord M:106 G:0 12 Extractor M:82 G:0 12 SpawningPool M:208 G:0 11 +1 Drone on gas M:23 G:0 12 +1 Drone on gas M:2 G:2 12 +1 Drone on gas M:2 G:2 17 Lair M:165 G:113 17 Hatchery M:303 G:70 18 Overlord M:106 G:111 18 ExtractorTrick M:85 G:127 20 Queen M:190 G:165 23 Spire M:204 G:201 23 Overlord M:158 G:35 24 +1 Drone on minerals M:184 G:63 25 Queen M:160 G:66 28 Queen M:162 G:87 30 Overlord M:100 G 6 33 Zergling M:157 G:124 34 Zergling M:107 G:124
After 325 million games the fastest way to get spire with 30 drones. But unfortunately gets the 4 zerglings right at the end, I should have used a way point with the 4 lings.
|
That does not look right at all to me. :-(
Early extractor and you only need 300 gas for this
You end up with 124 gas
Late queen and hatch and you need tones of larva for this.
What is the time on that? I am sure a way better BO exists
|
I cancelled the run after 350 million games because of the late zerglings. The timing was 6:08 for the last ling to pop (spire popped at 6:03).
|
On October 27 2010 19:49 redviper wrote:+ Show Spoiler +10 Overlord M:106 G:0 12 Extractor M:82 G:0 12 SpawningPool M:208 G:0 11 +1 Drone on gas M:23 G:0 12 +1 Drone on gas M:2 G:2 12 +1 Drone on gas M:2 G:2 17 Lair M:165 G:113 17 Hatchery M:303 G:70 18 Overlord M:106 G:111 18 ExtractorTrick M:85 G:127 20 Queen M:190 G:165 23 Spire M:204 G:201 23 Overlord M:158 G:35 24 +1 Drone on minerals M:184 G:63 25 Queen M:160 G:66 28 Queen M:162 G:87 30 Overlord M:100 G 6 33 Zergling M:157 G:124 34 Zergling M:107 G:124 After 325 million games the fastest way to get spire with 30 drones. But unfortunately gets the 4 zerglings right at the end, I should have used a way point with the 4 lings.
How soon do you want the lings?
I'm running it now to get 4 lings @ 3:30 and spire final.
|
|
|
|
|
|
|
|
EPT
