|
On November 30 2010 12:22 jdseemoreglass wrote:Show nested quote +On November 30 2010 12:15 CarbonTwelve wrote:On November 30 2010 11:55 Skrag wrote:
And this is why I have no faith in the optimizers for this sort of thing. It's very easy to demonstrate that any 10pool is easily surpassed economically. The faster queen simply does not make up for the fact that you're wasting larvae spawn time, and slowing everything else down.
As my 11Overpool example showed, it's not nearly as bad as people thing, but those builds *will* be at an economic disadvantage, and the fact that the optimizer says different doesn't make it so. Personally I'd like to see proof that it's 'easily surpassed economically' before dismissing the optimiser so readily. At the moment I don't have enough time to do any testing or calculations of my own, but I do have a lot of faith in the optimiser I've built, and if it finds that 10 pool 16 hatch allows it to achieve the result faster, I'm fairly confident that that is indeed the case, or at the very least it isn't a significant disadvantage. uh, hey guys... I've pointed you in the direction of tests and results several times now. We have significant data on 10 pool, overpool, 14 pool, 16 pool, 14 hatch, 16 hatch, and more... You can use this to compare to the results of the optimizers, unless you think there is a significant flaw in our testing methods. http://www.teamliquid.net/forum/viewmessage.php?topic_id=172481
Cool thread, must have missed your previous posts about it. Although as I've said before, I don't think the "most minerals mined by 6:00" paradigm is perfect, I still think it's good to get some objective, tightly controlled, scientific data on the subject. Yay science!
Hopefully, some creative TLers can come up with ways to objectively compare builds that factor in cannon rushes, 2 raxes, and worker/ebay/pylon blocks. It'd be tough though...so many variables.
|
On November 30 2010 12:50 CarbonTwelve wrote:Show nested quote +On November 30 2010 12:22 jdseemoreglass wrote:uh, hey guys... I've pointed you in the direction of tests and results several times now. We have significant data on 10 pool, overpool, 14 pool, 16 pool, 14 hatch, 16 hatch, and more... You can use this to compare to the results of the optimizers, unless you think there is a significant flaw in our testing methods. http://www.teamliquid.net/forum/viewmessage.php?topic_id=172481 Yeah, and this basically confirms that the economy of 11 pool 18 hatch is pretty close to 16 hatch 15 pool (based on there being slightly fewer minerals mined, but more drones meaning higher income, which should balanced out the minerals mined), which is what I was suggesting. That's why I say I'd like to see some proof from Skrag that 10 or 11 pool is 'easily surpassed economically'.
There's a *huge* problem in that thread, that is very likely causing some misleading info. While you can absolutely say that 11pool 18hatch is "pretty close" to 16hatch 15pool, in that 16hatch is ahead on minerals mined, you can't actually say for sure that the 11pool is truly ahead on larva. Sure, it's ahead at exactly 6 minutes, but here's the problem:
When I was testing 15hatch vs 16pool, 16pool had a significant lead in queen spawn time, but the hatch-first got a significant drone lead before the first larva spawn ever finished. The end result is that 15hatch was significantly ahead on average, but there were moments in time that you could pick, and it would look like the 16pool was ahead larvae-wise, because it had more available larva and more completed drones. But what was really happening is that hatch-first was 3 workers ahead for a period of time, then pool-first was ahead 1 worker for a period of time, then hatch-first was 3 workers ahead again. Depending on what point in time you chose to look at the game, you would "see" a different result, and it would look like pool-first was ahead, even though it was behind more than it was ahead (actually it was ahead for longer periods of time, but only ever 1 drone ahead)
Everybody in that thread is completely ignoring the spawn larvae cycle, and just saying "hey look, 11pool is some minerals behind the 16hatch, but is 3 drones ahead at point X" where point X just happens to be a point that's nearly ideal for the 11pool, because drones from the most recent 2 rounds of spawn larvae would just be finishing.
Look 20 seconds later, and it might look like 16hatch is ahead in both minerals *and* larva.
I haven't specifically done detailed testing between 11pool and 14/15/16 hatch, just the quickie results that I already posted earlier in this thread, which showed 11pool being significantly behind everything else I tested, in both larvae and minerals.
To be perfectly fair though, after reading that other thread, I tried to reproduce the OPs result with the 11pool, and had to adjust a number of things before I could even come close, so my quickie test might not be valid anyway. I didn't do any more preparation or practice on the other builds I was comparing it against, but it turns out there are some pretty tricky and counter-intuitive spots in an optimal 11overpool build.
However, I have done previous very specific testing that showed a 9OL 14pool was better at doing one particular zergling/nydus one-base all-in than than an 11 overpool build, where "better" was defined as having all the necessary units and upgrades at the best possible time, and the 14pool was better because it could build things sooner due to have a better resource base. That was a onebase build, though, so obviously hatchery timing never came into it.
There's another danger caused by the sort of testing that's going on in this and that other thread though. You are *never* building purely drones for 6 straight minutes. And that unrealistic situation actually causes difficulty executing some builds for the sake of comparison. For example, there are a few iterations of 14hatch or 15hatch where I really needed to be able to build a gas geyser, and it was at a point where I would be wanting to start getting gas in a real game anyway, but building the geyser and mining gas would invalidate the comparison, or at least make it much more difficult. But not building the geyser meant putting a huge speed bump in the build, having to build an overlord one larva sooner than was ideal for the build, which ends up delaying a queen on minerals, because of the extra 50 minerals spent on that particular larvae that really wanted to be a drone instead of an overlord. And if I chose to not delay the queen, the drones on the other end would be delayed. All that just because I couldn't build a gas geyser at a particular time. (or at all for that matter). A lot of really important timings don't actually become apparent until you actually start trying to build the buildings or units involved in them. I guess that's obvious, but any sort of thread like this needs to be really careful. Because we don't ever actually care about what the "most economic" build is. In fact, the phrase itself can mean different things to different people. What we truly care about is "what opening lets me best execute A, B, or C, because it gets X resources the fastest", which is a very different question, and one that can't be answered by a simple drone race. So far, the answer seems to be that if you need more resources than larva, hatch-first builds are going to be better pretty much always, but if you need more larva than resources, pool-first builds, including 11overpool, *might* be faster, but even that is unclear because it's difficult to truly measure the effect of the spawn larva cycles.
There's also a distortion effect caused by the fact that your bases are oversaturated far sooner than they would be in a real game where you were building an occasional building or military unit.
So while a simple drone race can give interesting information (as long as everything is taken into account, such as queen spawn larva cycles), if it goes on too long, the test itself can distort the results. 6 minutes is *probably* slightly too long if you're not going to mine any gas. 40 drones gets you to the point where internal randomness can have a big effect on mining rates, as workers try to settle into a routine rather than bouncing around looking for a spot to mine. 7 minutes is *definitely* too long. The results when comparing builds will be consistent if there's a big enough difference between them, but there can be a lot of variance from run to run of a single build.
And this is the point where this post has become so long, and it's become so late (early!) that I don't remember what my original purpose was, and I"m far too tired to go back and find out.
So umm...
I hope that made sense.
|
On November 30 2010 12:05 Skrag wrote:Show nested quote +On November 30 2010 10:01 icezar wrote:
You did something wrong, i just tested this exactly how you said and i get:
14pool 16hatch
734/368 60/60
2 drones 12/17
2 OL 9/25
Spawn 3/40 $ just finished
6 larva Ok, beyond the obvious questions (did you do it on lost temple at the 9 oclock position? 12oclock will give different results), it just doesn't matter if I did something wrong compared to you. I did things the same way for *myself* in every game, down to making sure I got the same split at the beginning (and restarting yabot if I misclicked), which still leaves a valid comparison between *my* 14pool 16hatch and *my* 15hatch 15pool. If you're slightly more optimal in doing everything than I am, that really just doesn't matter. Because whatever inefficiencies I have as a player will be consistent from game to game. Although to be perfectly honest, I'm really suspecting you played at 12-oclock or on a different map altogether, because I'm not 100% on the ball, but there's no way I'm *that* far off.
Yes i did Lost temple 9 oclock.
Today i also tested 15hatch 15pool, you can see that they are identical :-))))))) just a few more minerals but i think it is because 18 Extractor is later and you can see you also have less gas.
I have 1 done finished vs 1 drone 13/17 and you have 1 more larva
15hatch 15pool
750/340 60/60
3 drones 13/17
2 OL 10/25
Spawn 11/40 & Just finished
5 larva
I think what you did wrong in your test is you wait for your queen at expansion while i build my second queen asap on base an move my first queen to natural.
|
On November 30 2010 18:43 icezar wrote:
I think what you did wrong in your test is you wait for your queen at expansion while i build my second queen asap on base an move my first queen to natural.
I believe this is exactly right. I don't typically do pool-first fast expansions (if I pool first, it's to one base a while with a later expansion, and if I plan to FE, it's almost always hatch first), so I wasn't doing the obvious and building the second queen at the main.
My bad.
The numbers you got for 15hatch/15pool are almost exactly what I got.
Still, though, what you're showing is that 14pool/16hatch is *equal* to 15hatch/15pool (and after some more testing, I suspect 14hatch/14pool will do slightly better), where the original claim is that it was better.
|
On November 30 2010 18:51 Skrag wrote:
Still, though, what you're showing is that 14pool/16hatch is *equal* to 15hatch/15pool (and after some more testing, I suspect 14hatch/14pool will do slightly better), where the original claim is that it was better.
While it might not necessarily be 'better' in terms of economy, you have to admit that it's certainly a LOT safer due to having the pool out much earlier. Given relatively equal economies I can't see any significant advantage for hatch before pool. The only one I can think of is avoiding being blocked, but then having the pool out earlier would deal with that anyway.
|
On November 30 2010 20:00 CarbonTwelve wrote:Show nested quote +On November 30 2010 18:51 Skrag wrote:
Still, though, what you're showing is that 14pool/16hatch is *equal* to 15hatch/15pool (and after some more testing, I suspect 14hatch/14pool will do slightly better), where the original claim is that it was better. While it might not necessarily be 'better' in terms of economy, you have to admit that it's certainly a LOT safer due to having the pool out much earlier. Given relatively equal economies I can't see any significant advantage for hatch before pool. The only one I can think of is avoiding being blocked, but then having the pool out earlier would deal with that anyway.
I do not agree totally. Hatch first has more larvae or at least faster at the beginning this helps you to hold any attacks.
11 Pool is safer only until 3:40 min but this is helpful only in ZvZ i think. For T and P to hit you with something before that i think they will just die after you build only zerglings from 2 hatch.
So hatch first, faster larva, faster creep, and a lot harder to block. You just have to learn how to protect your ramp from pylons and bunkers using drones.
But the difference is very small so if you feel better with pool first it is nothing wrong.
I think the problem with your app is that it is not real enough, if you test a build from it you cannot do it exactly, you will always be a few seconds behind, and Hatch first does not require drone transfer, with pool first you have to transfer some drones and this is not taking into consideration.
Edit: a good example is 9 OV, your app alawys thinks 10 OV ExtrTrick is better but if you test it in game it is not if you pool later than 13
|
On November 30 2010 18:40 Skrag wrote:
What we truly care about is "what opening lets me best execute A, B, or C, because it gets X resources the fastest", which is a very different question, and one that can't be answered by a simple drone race.
What you say is very true. Yet one of the important results from these threads is myth-busting the idea that hatch first is the only eco build. Last week, an 11-pool would have been criticized as a terrible eco build, and labeled as "all-in". I expect to see some new dynamic BOs as a result.
|
On November 30 2010 21:51 icezar wrote:
I do not agree totally. Hatch first has more larvae or at least faster at the beginning this helps you to hold any attacks.
I'm not sure that that's true though. Looking at the two builds in the other thread (http://www.teamliquid.net/forum/viewmessage.php?topic_id=172481), going hatch first gives you the second hatchery at 4 mins, but if you go 11 pool then the queen finishes at 3:20, meaning your first spawn larvae will pop at 4 mins also.
On November 30 2010 21:51 icezar wrote:
I think the problem with your app is that it is not real enough, if you test a build from it you cannot do it exactly, you will always be a few seconds behind, and Hatch first does not require drone transfer, with pool first you have to transfer some drones and this is not taking into consideration.
Edit: a good example is 9 OV, your app alawys thinks 10 OV ExtrTrick is better but if you test it in game it is not if you pool later than 13
I believe these are just implementation details - slight alterations in the times required for these should fix up the algorithm so that it replicates reality a lot closer. I totally agree that's it's not perfect atm, but I think it can get better, so I don't believe that optimisers should be dismissed so readily based on their current state.
|
this thread is full of false theorycrafting.
skrag pretty much sums it up but i have another point to add.
you are constantly building drones with your economy builds. this is just plain so wrong. just look at fruitdealers gsl match today. if he had build drones constantly he would have lost. you have to SAFE larvae to be able to react to that terran/protoss push.
While it might not necessarily be 'better' in terms of economy, you have to admit that it's certainly a LOT safer due to having the pool out much earlier. Given relatively equal economies I can't see any significant advantage for hatch before pool. The only one I can think of is avoiding being blocked, but then having the pool out earlier would deal with that anyway.
show me one game of a pro doing your 16 pool 15 hatch that is soooo superior to 14/15 hatches. you won't find any whereas you see the hatch first regularly. maybe because 16 pool 15 hatch would lose to every early timing push because you have to build the drones.
to the 11 pool builds: if you plan on 11pooling and just building drones you are SO vulnerable to early rushes, additionally to the potential early drones you sacrifice while saving 200 minerals at 11 supply. again: you have to safe larvae at some point until you know what your opponent is up to or you will lose because you don't have the production capabilitys to react to early pushes (fe 2 rax...).
|
On November 27 2010 08:56 Steven.Bonnell.II wrote:
Pool first is 100% better than hatch first in every single scenario. Since there are pro gamers that hatch first, many people will mindlessly copy without question.
If you run a side-by-side comparison and race to 50 drones, you'll find that hatch first and pool first both reach it at about the same time. The only difference is, pool first grants you much more flexibility and is a safer opening build.
There is absolutely no reason to hatch first in any matchup, on any map, ever.
Wow, this makes me take notice at least, as I remember you going fast expo every single game some times back on your stream.
|
On December 14 2010 21:03 Ghad wrote:Show nested quote +On November 27 2010 08:56 Steven.Bonnell.II wrote:
Pool first is 100% better than hatch first in every single scenario. Since there are pro gamers that hatch first, many people will mindlessly copy without question.
If you run a side-by-side comparison and race to 50 drones, you'll find that hatch first and pool first both reach it at about the same time. The only difference is, pool first grants you much more flexibility and is a safer opening build.
There is absolutely no reason to hatch first in any matchup, on any map, ever. Wow, this makes me take notice at least, as I remember you going fast expo every single game some times back on your stream.
It's not actually accurate what he said, more extensive testing has been done and hatch first is indeed slightly more economical (also, more importantly, it's impossible to have enough larva to fend off the 2-racks and come out close to even with a pool-first).
This topic was a bit old, I suggest you have a look at the topic here that focuses on finding the #1 highest eco build, while other topics like this one and another I've seen are just looking for "best" which is why I think there is so much argument: there's no such thing.
http://www.teamliquid.net/forum/viewmessage.php?topic_id=174374
|
And the results are pretty much in. 14h15p and 15h14p are pretty much equivalent economically, with differences so small that your mechanics are going to make a 10x bigger difference anyway. 13h15p appears to be only slightly behind, with a slight larva advantage, and 14h14p is a little bit further behind (but still a really small amount), with a larger larva advantage.
Those hatch-first builds are pretty clearly ahead of every pool-first build we tried, and are also ahead of later hatch-first builds, such as 16h15p, so those 4 builds are pretty much the cream of the crop economically.
For pool-first builds, 13h15p seems to be a pretty clear winner, and has a larva advantage over every other build tested, including the hatch-first builds. But it does come about 100 minerals short of the hatch first builds. Of the pool-first builds that were tested, 11p18h was the loser economically, but not by much.
So yes, hatch first really is more economical, although an early queen can go quite a long way in making up some of that difference, as the 11pool build shows.
|
|
|
|
|
|
EPT
