|
On November 09 2010 04:36 CtrlAltDefeat wrote:
I think it would be very helpful to release some kind of help document for explaining the UI in more detail.
More specifically, what exactly does "Enforce Worker Parity" do?
Enforce worker parity means it forces the build order to maintain the same worker count as a Terran player would in the same time.
On a side note, would it be possible to enforce income parity? You don't need to have as many workers if you're on 2 base and they're on 1. Might be a useful feature.
|
On November 09 2010 05:49 Goobus wrote:Show nested quote +On November 09 2010 04:36 CtrlAltDefeat wrote:
I think it would be very helpful to release some kind of help document for explaining the UI in more detail.
More specifically, what exactly does "Enforce Worker Parity" do? Enforce worker parity means it forces the build order to maintain the same worker count as a Terran player would in the same time. On a side note, would it be possible to enforce income parity? You don't need to have as many workers if you're on 2 base and they're on 1. Might be a useful feature.
Yes, but a Terran player on how many bases?
|
On November 09 2010 06:25 CtrlAltDefeat wrote:Show nested quote +On November 09 2010 05:49 Goobus wrote:On November 09 2010 04:36 CtrlAltDefeat wrote:
I think it would be very helpful to release some kind of help document for explaining the UI in more detail.
More specifically, what exactly does "Enforce Worker Parity" do? Enforce worker parity means it forces the build order to maintain the same worker count as a Terran player would in the same time. On a side note, would it be possible to enforce income parity? You don't need to have as many workers if you're on 2 base and they're on 1. Might be a useful feature. Yes, but a Terran player on how many bases?
A terran on the same number of bases as the BO being calculated.
|
Ive not read through all 50 pages yet, but is a terran or toss version of this program in the works?
|
I fixed the missing buttons on smaller resolutions. It will be in the next version. (so stop reporting this bug  )
|
|
|
Sweet. Soon we will be able to counter the 7RR with some sort of horrible 12 zealots at 3:30 attack.
|
On November 09 2010 08:54 DiracMonopole wrote:
Sweet. Soon we will be able to counter the 7RR with some sort of horrible 12 zealots at 3:30 attack.
If you think the counter to early roaches is early zealots, a build order optimizer isn't going to help you 
|
On November 09 2010 09:57 Bumblebees wrote:Show nested quote +On November 09 2010 08:54 DiracMonopole wrote:
Sweet. Soon we will be able to counter the 7RR with some sort of horrible 12 zealots at 3:30 attack. If you think the counter to early roaches is early zealots, a build order optimizer isn't going to help you
If you could actually get a dozen zealots at 3:30, yes, it would defeat early roaches.
|
On November 09 2010 11:39 DiracMonopole wrote:Show nested quote +On November 09 2010 09:57 Bumblebees wrote:On November 09 2010 08:54 DiracMonopole wrote:
Sweet. Soon we will be able to counter the 7RR with some sort of horrible 12 zealots at 3:30 attack. If you think the counter to early roaches is early zealots, a build order optimizer isn't going to help you If you could actually get a dozen zealots at 3:30, yes, it would defeat early roaches.
Im pretty sure 7 roachs can micro damn near infinite zealots in groups of 12. I wouldn't expect the BO finder to find anything that useful for toss or terran, much in the same way this BO calculator is kinda useless beyond 50 food. Pro level players have already figured out the best things to do, this roach 'find' isn't really a find. Its pretty damn logical to extrator trick/overpool into roaches and make as many as you can and then drone. Put in 30/30/30 of the base units and some basic upgrades, let it run for 1.5 billion games and it tells me to 14 pool/15 hatch. Thanks. Can it also link me to Idra's replays?
This is a great tool, and fun to play around with. But revolutionary? I think not.
|
On November 09 2010 14:05 Roaming wrote:Show nested quote +On November 09 2010 11:39 DiracMonopole wrote:On November 09 2010 09:57 Bumblebees wrote:On November 09 2010 08:54 DiracMonopole wrote:
Sweet. Soon we will be able to counter the 7RR with some sort of horrible 12 zealots at 3:30 attack. If you think the counter to early roaches is early zealots, a build order optimizer isn't going to help you If you could actually get a dozen zealots at 3:30, yes, it would defeat early roaches. Im pretty sure 7 roachs can micro damn near infinite zealots in groups of 12.
i'm pretty sure that the 7 roach rush pops the 7 roaches at 4:40 and has 0 defense at 3:30
I wouldn't expect the BO finder to find anything that useful for toss or terran, much in the same way this BO calculator is kinda useless beyond 50 food. Pro level players have already figured out the best things to do, this roach 'find' isn't really a find. Its pretty damn logical to extrator trick/overpool into roaches and make as many as you can and then drone. Put in 30/30/30 of the base units and some basic upgrades, let it run for 1.5 billion games and it tells me to 14 pool/15 hatch. Thanks. Can it also link me to Idra's replays?
This is a great tool, and fun to play around with. But revolutionary? I think not.
It found a very good timing push for zerg noone has yet thought of.
It will be easily possible to find perfect builds for terran and protoss up to much more than 50 food, since they produce units all the time, though have their defences set up. Unlike zerg, which would at the very best make every unit to defend last-second. (That's why +40-50 food builds do not work for zerg)
It may not be revolutionary, but I bet it can optimize some "pro" builds.
|
It may not be revolutionary, but I bet it can optimize some "pro" builds.
I do not think that at all!
I try to put in the 5RR and optimize it, it fails miserably, after a lot of hours and a lot of #K games played it is not even close!!!
What i find it useful is to give you a general build order that you can fine tune.
And a lot of fun for sure!
|
On November 10 2010 01:14 icezar wrote:I do not think that at all! I try to put in the 5RR and optimize it, it fails miserably, after a lot of hours and a lot of #K games played it is not even close!!! What i find it useful is to give you a general build order that you can fine tune. And a lot of fun for sure!
How many games exactly? People in this thread have been reporting changes to their builds even after 1.3 billion games. It's really something you want to leave running overnight to use it to full effect.
|
The new version doesn't seem to be evolving to the 7:32 7RR build for me whereas the v0017 version finds it within minutes. v0020 seems to get stuck at a 4:40 build without a queen. Did something change in terms of timings in v0020 ?
|
Not sure if I should be putting this into a new thread or just posting here, but basically after being inspired by Lomilar's app I was keen to try it myself (I've always been interested in AI and puzzle solvers), so I wrote my own GA in C++ with a lot of efficiency work I've done in other projects. The results are pretty good, and because of the efficiency of C++ and the custom memory management class I've got it's able to process 500,000+ games per second. I've also written a Protoss module (actually I started with that - just getting Zerg done now), although that does run a bit slower due to the handling for Chrono Boost. I also made a few changes to the way the GA works, implementing a large city that has immigration from several small villages.
The code isn't quite ready for public release yet (mainly because it doesn't have a UI yet), and there are still a few features that need to be implemented, but I thought I'd post here and get people's thoughts on it and see if there's much interest from others in pursuing a design under this architecture.
Note, I really don't want to take any credit away from Lomilar - I did PM him asking if he was interested in looking at this design, but I haven't heard back so I thought I'd just post in this thread. Anyway, it was his idea, all I've done is just implement it with a more efficient architecture.
Also, if you've got any Protoss builds you want me to test out for you and post the results, just let me know.
- Carbon
|
I feel this tool is slightly overrated.
Just play the damn game and you'll get better. Although your ladder rank may look a bit shite, trying it against real players is better than this IMO
|
On November 10 2010 07:52 CarbonTwelve wrote:
Not sure if I should be putting this into a new thread or just posting here, but basically after being inspired by Lomilar's app I was keen to try it myself (I've always been interested in AI and puzzle solvers), so I wrote my own GA in C++ with a lot of efficiency work I've done in other projects. The results are pretty good, and because of the efficiency of C++ and the custom memory management class I've got it's able to process 500,000+ games per second. I've also written a Protoss module (actually I started with that - just getting Zerg done now), although that does run a bit slower due to the handling for Chrono Boost. I also made a few changes to the way the GA works, implementing a large city that has immigration from several small villages.
The code isn't quite ready for public release yet (mainly because it doesn't have a UI yet), and there are still a few features that need to be implemented, but I thought I'd post here and get people's thoughts on it and see if there's much interest from others in pursuing a design under this architecture.
Note, I really don't want to take any credit away from Lomilar - I did PM him asking if he was interested in looking at this design, but I haven't heard back so I thought I'd just post in this thread. Anyway, it was his idea, all I've done is just implement it with a more efficient architecture.
Also, if you've got any Protoss builds you want me to test out for you and post the results, just let me know.
- Carbon
Wow sounds amazing! About 10,000 games a second is the most I can get out of a reasonable complex build. 500,000 a second is a huge performance increase!
|
On November 10 2010 07:52 CarbonTwelve wrote:
Not sure if I should be putting this into a new thread or just posting here, but basically after being inspired by Lomilar's app I was keen to try it myself (I've always been interested in AI and puzzle solvers), so I wrote my own GA in C++ with a lot of efficiency work I've done in other projects. The results are pretty good, and because of the efficiency of C++ and the custom memory management class I've got it's able to process 500,000+ games per second. I've also written a Protoss module (actually I started with that - just getting Zerg done now), although that does run a bit slower due to the handling for Chrono Boost. I also made a few changes to the way the GA works, implementing a large city that has immigration from several small villages.
The code isn't quite ready for public release yet (mainly because it doesn't have a UI yet), and there are still a few features that need to be implemented, but I thought I'd post here and get people's thoughts on it and see if there's much interest from others in pursuing a design under this architecture.
Note, I really don't want to take any credit away from Lomilar - I did PM him asking if he was interested in looking at this design, but I haven't heard back so I thought I'd just post in this thread. Anyway, it was his idea, all I've done is just implement it with a more efficient architecture.
Also, if you've got any Protoss builds you want me to test out for you and post the results, just let me know.
- Carbon
How about fastest 7 stalkers?
|
On November 10 2010 09:51 Dionyseus wrote:
How about fastest 7 stalkers?
For some reason it never occurred to me to try that. I left it running over my lunch (~45 mins) and here's what it comes up with:
+ Show Spoiler +Evolution: 185500
Games played: 840271182
Games per second: 370067
0.00: 50M 0G 0E 6/ 10S - Build Probe
17.00: 73M 0G 10E 7/ 10S - Build Probe
34.00: 107M 0G 19E 8/ 10S - Build Probe
41.63: 100M 0G 23E 9/ 10S - Build Pylon
51.65: 50M 0G 29E 9/ 10S - Build Probe
68.65: 95M 0G 39E 10/ 18S - Build Probe
68.65: 45M 0G 39E 11/ 18S - Chrono Nexus
72.85: 75M 0G 16E 11/ 18S - Build Assimilator
82.25: 60M 0G 21E 11/ 18S - Build Probe
97.65: 131M 0G 30E 12/ 18S - Build Probe
102.85: 125M 0G 33E 13/ 18S - Move Probe To Gas
105.99: 150M 2G 35E 13/ 18S - Build Gateway
105.99: 0M 2G 35E 13/ 18S - Chrono Nexus
113.03: 50M 7G 14E 13/ 18S - Build Probe
113.03: 0M 7G 14E 14/ 18S - Move Probe To Gas
126.78: 98M 26G 21E 14/ 18S - Build Probe
126.78: 48M 26G 21E 15/ 18S - Move Probe To Gas
143.78: 181M 59G 31E 15/ 18S - Build Probe
160.78: 277M 91G 40E 16/ 18S - Build Probe
160.78: 227M 91G 40E 17/ 18S - Build Gateway
170.99: 165M 110G 46E 17/ 18S - Build Cybernetics Core
186.05: 150M 139G 55E 17/ 18S - Build Gateway
196.86: 100M 160G 61E 17/ 18S - Build Pylon
207.67: 100M 180G 67E 17/ 18S - Build Pylon
221.86: 131M 207G 75E 17/ 26S - Build Stalker
221.86: 6M 157G 75E 19/ 26S - Chrono Gateway
233.86: 125M 180G 57E 19/ 34S - Build Stalker
251.05: 170M 162G 66E 21/ 34S - Build Stalker
251.05: 45M 112G 66E 23/ 34S - Chrono Gateway
259.11: 125M 128G 46E 23/ 34S - Build Stalker
259.11: 0M 78G 46E 25/ 34S - Chrono Gateway
275.86: 166M 110G 30E 25/ 34S - Build Stalker
288.05: 162M 83G 37E 27/ 34S - Build Stalker
297.11: 126M 50G 42E 29/ 34S - Build Stalker
297.11: 1M 0G 42E 31/ 34S - Chrono Gateway
302.04: 50M 9G 20E 31/ 34S - Build Probe
311.11: 90M 27G 25E 32/ 34S - Chrono Gateway
330.71: 291M 64G 11E 32/ 34S - Satisfied.
The reason it builds that probe right at the end is because the fitness algorithm I'm using tries to maximise its value including stuff you haven't requested, so it builds another probe to get a bit of extra mining done in the last few seconds.
Also, the functions I'm using for calculating income rates are a little more accurate than the ones in Lomilar's app (well, more accurate compared to the measurements as calculated by Piousflea: http://www.teamliquid.net/forum/viewmessage.php?topic_id=140055), so this will explain any slight differences you might notice in available resources if you compare the two.
|
sounds good, CarbonTwelve. I also wanted to write one of these in C++. Perhaps I can help with yours, if you like and if I find the time.
PM me if you're interested.
|
|
|
|
|
|
EPT
