|
I must admit from what i see of the 11 overpool 18 hatch build, i'm very impressed.
I'd always imagined that the "standard" openers Z uses weren't actually as "greedy" or econ focused as people (including me) believed. The addition of a much earlier pool makes Z so much more flexible and safe early game too, or leads into an insane econ.
Whilst hatch first builds seem more intuitively econ focused, and despite that fact that "all the pros do it" I had always hoped somebody would actually just do the maths lol.
I'm far too lazy and uneducated for it, but I'm glad somebody is going for it  .
|
jdseemoreglass: with regard to how you should weigh the 'more drones but less minerals mined' or 'less drones but more minerals mined' - speaking from what I would believe is a practical standpoint, I would value the higher drone count more. After all, you *are* going to be making structures with your drones. Extractors. Possibly a roach warren. Mining the gas from those extractors. If your build already gives you less drones, then I would imagine that when you need to sacrifice drones in order to make those structures, you're losing more minerals than the build with the higher drone count and thus being economically suboptimal when compared with the high drone count build.
What do you think? I haven't actually done the testing, but it seems to follow intuitively.
|
I agree with Enyalus.. more drones = more buildings = less money spent on drones = more money to spent on other things, such as drones.
|
@B34ST
I dont quite agree. I think its rare that your ONLY going to be making drones, so some of your larva is going to be spent on lings/roaches. In that case, I would say having more minerals is more important, as then you got more to place structures/build lings, while still keeping up a good drone count. Having more drones (and less total minerals mined), means you will need to cut drones to get enough minerals to build something like a roach warren. Of corse the counter argument is you might have to cut larva to get lings/roaches out in the more mined build, but in that case I would think you would just get an extra queen earlier for more larva.
Still its probably close enough these two are practicly tied.
|
On November 29 2010 05:26 jdseemoreglass wrote:Show nested quote +On November 29 2010 04:33 Hurkyl wrote:
Also, it is relevant to note where in the larvae cycle the hatcheries are, as well as the spawn larvae. There's a big difference between having the next spawned larvae drop at 6:01 and 6:39! This is an important point that I'm glad someone mentioned... However, I feel these numbers are less relevant toward overall economy than drone count (including production) and minerals mined. If two builds are very similar in these regards, then the "larva in production" data would also be factored in.
The spawn larvae cycle is actually *more* important than the current drone count. For example, at 6 minutes, your "winning" build at 43 drones has the benefit of finishing drones off 2 recent queen spawns, while the "inferior" builds that I've tried, and the 16h/15p replay, only have the benefit of 1. Shift the recording time 20 seconds or so, and suddenly things would look different.
The only reason 11Overpool is even ahead at that point in time is because of the earlier first queen. Hatch-first builds make up for that by having extra larvae sooner from the hatch, and by having a slightly faster second queen. Ignoring the cycle means your measurements are flat out incorrect, because the two builds will flip-flop in drone count, with whichever one most recently finished a queen spawn being ahead. If hatch-first is ahead more often than it's behind, or ahead by more when it is ahead, that tips the balance in its favor both in larva count *and* resource count.
Also, I'm fairly sure 14hatch/14pool is better than 16hatch/15pool. It seemed to be in my testing, but I wasn't able to totally match the times and resource counts showed in your replays.
|
On November 30 2010 15:17 Enyalus wrote:
jdseemoreglass: with regard to how you should weigh the 'more drones but less minerals mined' or 'less drones but more minerals mined' - speaking from what I would believe is a practical standpoint, I would value the higher drone count more. After all, you *are* going to be making structures with your drones. Extractors. Possibly a roach warren. Mining the gas from those extractors. If your build already gives you less drones, then I would imagine that when you need to sacrifice drones in order to make those structures, you're losing more minerals than the build with the higher drone count and thus being economically suboptimal when compared with the high drone count build.
What do you think? I haven't actually done the testing, but it seems to follow intuitively.
You still have to make all those structures if you have the higher drone count. 
It takes time before the higher Drone count makes up for the smaller initial mineral count, and this advantage vanishes entirely if your opening doesn't have enough income to continue using Larvae at full capacity.
The main advantage to a higher Larvae count, I think, is for builds that switch over to making combat units for a while, because it lets you get the same army with more Drones (or a bigger army if you prefer).
|
On November 29 2010 05:49 Voros wrote:
Using this methodology rather than optimizers (what you dismiss as "theorycrafting") introduces human error into the equation. Your initiative to employ the scientific method is admirable, but the reality is that reductionistic approaches (i.e., the genetic algorithm that has become so popular these days) are going to be far more accurate in this very limited sphere than any method whose results depend on perfectly consistent performance from both the player as well as the network.
You mean except for the fact that those genetic algorithms fail to give the correct answers for anybody who's not 100% perfect with their timing, whose workers gain minerals in chunks of 5 instead of at a constant rate, and whose workers actually have to move to create buildings?
Oh wait. That's pretty much everybody in the world.
Optimizers can give you a head start, or give you ideas, or even give you the foundation of a build, but they cannot actually tell you the fastest way to get to a particular point, because they're not based on the reality of a person playing the game, or even the reality of the game itself. They take shortcuts so that the problem can actually be solved in a reasonable amount of time, and without rewriting and perfecting starcraft's game simulation engine.
|
With regards to your dilemma in the most recent build, the answer is simple: most minerals wins.
More drones is great, but you clearly defined "winning" as "most minerals mined" with other things have very little extra weight.
The fastest way to 2-base saturation and the fastest way to 2-base 2/3 saturation are likely to be different.... just as the most economical build from the 5 minute mark is likely to be different from this.
If you want to be extra cool, graph the total resources collected as a function of time with all of these builds, and then for any given time, you'll have the optimal build.
What's the purpose of a pretty graph?
Whenever you end up building beyond just a pool, you'll know which opener is the most economical for you. If you want to make a warren, push at 5 minutes, etc etc etc ... You'll know what opener will help you to transition into the rest of the opening while maintaining the most econ until that switch happens.
|
On November 30 2010 15:55 Skrag wrote:Show nested quote +On November 29 2010 05:49 Voros wrote:
Using this methodology rather than optimizers (what you dismiss as "theorycrafting") introduces human error into the equation. Your initiative to employ the scientific method is admirable, but the reality is that reductionistic approaches (i.e., the genetic algorithm that has become so popular these days) are going to be far more accurate in this very limited sphere than any method whose results depend on perfectly consistent performance from both the player as well as the network. You mean except for the fact that those genetic algorithms fail to give the correct answers for anybody who's not 100% perfect with their timing, whose workers gain minerals in chunks of 5 instead of at a constant rate, and whose workers actually have to move to create buildings? Oh wait. That's pretty much everybody in the world. Optimizers can give you a head start, or give you ideas, or even give you the foundation of a build, but they cannot actually tell you the fastest way to get to a particular point, because they're not based on the reality of a person playing the game, or even the reality of the game itself. They take shortcuts so that the problem can actually be solved in a reasonable amount of time, and without rewriting and perfecting starcraft's game simulation engine.
Two corrections:
I think you're talking about "simulators" not "optimizers". Something that takes a build order and reports timings, resources, and so forth is a simulator. An optimizer is something that tries to produce the best build order according to some criterion. I imagine the confusion has arisen because optimizer needs to have access, and people are more familiar with the optimizer.
Secondly, if such accuracy is desired, it is not all that difficult to make a good simulation that takes into account things like the positioning of mineral patches, transit time of peons for mining and building, imperfect peon splitting in the first second, and so forth. I imagine the only reason the existing simulators do not do this is because the authors judged the extra accuracy was not worth the programming and modeling effort.
|
Skrag,
The overpool means that you get a whole larva spawn for free, albeit on one base, because it advances that queen to 16 without having to wait for a hatch's minerals, instead of doing the queen at 18/ish after hatch. It also utilizes the queen build portion of the hatchery, which is, nonintuitively, a production facility.
Hatch before pool still doesn't get you a queen early, though it does get you about 3 larva (which is why the builds are so close)
Also, queen pops 4 larva in 40 seconds (+5 for re-energy) vs the hatch which only spawns 3 in 45 seconds. And costs half as much.
Also, because you start your second queen at 21 (most hatch-ish builds start around then too), the overpool is roughly on the same larva spawn timing as later builds.
(Note: I know supply numbers are not equal to time, but you get the idea).
Jd,
Adding an extractor trick was a good call. Seems obvious now. 11 drones then overpool > 10 drones then overpool. Probably don't waste any larva spawning time either.
I wonder if double trick overpool is teh ultimate!
Also, as to your question, larva are just as important to the economy as are minerals. A drone at the minerals is worth two in the bank.
Going to go see if I can re-modify the build. :-)
|
Here is how I would compare the 2 builds.
Take a first approximation that both builds at 6:00 are now able to make drones at the same rate, then the net drone difference remains fixed for the rest of the game.
At 6:00 minutes...
Lomilar is ahead by 2.58 drones, and by approximation will remain 2.58 drones ahead. (This is actually pretty fair, Lomilar is ahead on queen energy) However Lomilar is behind by 155 minerals.
So its 155 minerals vs 2.58 drones.
As long as Lomilar stays 2.58 drones ahead, he should catch up on the mineral metric, in ~1.5 minutes.
Ultimately this is a flaw in your claim here:
The time will be stopped precisely at 6 minutes for data to be assessed. Note that this time is not biased, as ALL data will be assessed in making a determination.
Based on this approximation, I would predict by 7.5-8 minutes Lomilar would have the most minerals mined, and thus the more economic opener.
The actual test would then be to actually do this, have fun someone
|
So, 11 pool 18 hatch.. impressive!
On November 29 2010 06:27 Lomilar wrote:
Edit: The retarded thing is, this is almost the 7RR opening, making drones/expansion instead of double-OL and roach warren around 18. Tricky!
Which makes it even more awesome because you can both be super economical and cheesy
This thread may become gamechanging
|
Tie Breaking:
Honestly, I think that taking a silly reductive approach "drone is worth x minerals * time" is a sort of silly reductionist approach to the problem you've spent all this time being empirical about.
The real answer is this:
Instead of having a single finish line at 6 minutes, what you should do is have 3 finish lines at 5, 6, and 7. The build that performs best across multiple time periods should be the winner (as in real game, it would allow for the most flexibility while still maxing out economy).
I would just compare raw minerals mined at each time point, absolutely nothing else (if needed add a 7:30 finish line, where no more drones are built after 7 - only mining is occurring in order to allow last minute drones to demonstrate value).
|
On November 30 2010 16:02 Hurkyl wrote:
I think you're talking about "simulators" not "optimizers". Something that takes a build order and reports timings, resources, and so forth is a simulator. An optimizer is something that tries to produce the best build order according to some criterion. I imagine the confusion has arisen because optimizer needs to have access, and people are more familiar with the optimizer.
I'm perfectly aware of the difference between simulators and optimizers.
My point was that for an optimizer to give correct results, it would have to have an accurate simulation embedded into it. (You are aware that an optimizer does actually have to simulate the build somehow in order to compare modifications, right?)
Secondly, if such accuracy is desired, it is not all that difficult to make a good simulation that takes into account things like the positioning of mineral patches, transit time of peons for mining and building, imperfect peon splitting in the first second, and so forth. I imagine the only reason the existing simulators do not do this is because the authors judged the extra accuracy was not worth the programming and modeling effort.
The current batch of optimizers give answers that are flat out *wrong*, and provably so, because they don't take reality into account, and quite frequently will even give out timings that simply don't work out due to the simplified (necessarily so, I'll admit, because accurately reproducing the game simulation would be an absolutely ridiculous amount of work) nature of their internal simulations.
Like I said, they can be good starting points, but actual in-game testing is required, and I jumped into this part of the conversation purely because somebody had the audacity to claim that you really shouldn't rely on in-game testing, because players are fallible, and the optimizers are supposedly not.
Except that they are, in fact, quite fallible. Extremely so. Not to the point that they're useless, but certainly to the point that in-game testing is *required* to verify the results of the optimizers, meaning that discouraging people from in-game testing in favor of these genetic algorithms is flat out retarded.
|
On November 30 2010 16:10 tehV wrote:So its 155 minerals vs 2.58 drones.
As long as Lomilar stays 2.58 drones ahead, he should catch up on the mineral metric, in ~1.5 minutes.
You are assuming infinitely many mineral patches. At 6:00 both builds are nearly totally saturated, the next few Drones will barely increase the mining rate. After 48 drones, remaining 2.58 drones ahead is a total waste of 129 minerals. (Also, the "Drones" line in the opening post is not a good estimate of how many Drones ahead each build will be, on average)
|
On November 30 2010 16:16 30to1 wrote:
The real answer is this:
Instead of having a single finish line at 6 minutes, what you should do is have 3 finish lines at 5, 6, and 7. The build that performs best across multiple time periods should be the winner (as in real game, it would allow for the most flexibility while still maxing out economy).
I would just compare raw minerals mined at each time point, absolutely nothing else
I find your finish lines completely arbitrary that, why not finish lines at 1, 2, 3, and/or 4 also? The OP wants the best economic opener, not the best economic opener that is also the best at 5,6,and7 minutes.
And stopping around 7 minutes is obviously to short as Lomilar's drone advantage wont show up till closer to 8 minutes.
His 6 minute mark is obviously just a simplification to help test the best economic opener. If you wanted to remove this time limit nonsense, all you need to do is redefine the problem as the best economic opener is the one that fully mines out a specific map the fastest. Have fun doing that manually.
Going to 6 minutes, noting the drone differences, etc.. and extrapolating from there is probably 'good' enough.
|
On November 30 2010 16:24 Hurkyl wrote:Show nested quote +On November 30 2010 16:10 tehV wrote:So its 155 minerals vs 2.58 drones.
As long as Lomilar stays 2.58 drones ahead, he should catch up on the mineral metric, in ~1.5 minutes. You are assuming infinitely many mineral patches. At 6:00 both builds are nearly totally saturated, the next few Drones will barely increase the mining rate. After 48 drones, remaining 2.58 drones ahead is a total waste of 129 minerals. (Also, the "Drones" line in the opening post is not a good estimate of how many Drones ahead each build will be, on average)
We don't need to assume infinitely many mineral patches, but zergs like to get a 3rd base right? Do zergs stop naturally in game at 48 drones?
Lomilar's drones will be an economic advantage, I don't see how you get around that.
|
Is anyone actually considering Douillos's build when Lomilar's gets you a pool in little over half the time, two extra drones with a third on the way, and all for just 150 minerals less?
It looks like I'm going 11 overpool every game from this point on .
|
On November 30 2010 16:30 tehV wrote:Show nested quote +On November 30 2010 16:16 30to1 wrote:
The real answer is this:
Instead of having a single finish line at 6 minutes, what you should do is have 3 finish lines at 5, 6, and 7. The build that performs best across multiple time periods should be the winner (as in real game, it would allow for the most flexibility while still maxing out economy).
I would just compare raw minerals mined at each time point, absolutely nothing else
I find your finish lines completely arbitrary that, why not finish lines at 1, 2, 3, and/or 4 also? The OP wants the best economic opener, not the best economic opener that is also the best at 5,6,and7 minutes. And stopping around 7 minutes is obviously to short as Lomilar's drone advantage wont show up till closer to 8 minutes. His 6 minute mark is obviously just a simplification to help test the best economic opener. If you wanted to remove this time limit nonsense, all you need to do is redefine the problem as the best economic opener is the one that fully mines out a specific map the fastest. Have fun doing that manually. Going to 6 minutes, noting the drone differences, etc.. and extrapolating from there is probably 'good' enough.
If you limit yourself to mining out two bases, you don't have to do anything with "drone differences". At 7:00 or 7:30 or so, all builds considered will have 48 Drones, which (I believe) gives the maximum mining rate. So you just have note how many minerals are mined at that point in time.
Well, I guess this approach ignores the effects of when the mineral patches vanish. (but so does what you suggest). It could be accounted for, though, if anyone really cared.
|
On November 30 2010 16:08 Lomilar wrote:
Skrag,
The overpool means that you get a whole larva spawn for free, albeit on one base, because it advances that queen to 16 without having to wait for a hatch's minerals, instead of doing the queen at 18/ish after hatch. It also utilizes the queen build portion of the hatchery, which is, nonintuitively, a production facility.
Here's the thing though. Until that first spawn finishes, the hatch-first build is *always* ahead, by a number of workers that depends on when the hatch and pool specifically went down. Being ahead early is better than being ahead later, unless it's a *lot* ahead later. But once the hatch-first builds start their spawn cycles, the pool-first builds are typically doing catch-up from the early lead, and maybe spending a small amount of time during the cycle being ahead, and as I mentioned, the lead flip-flops based on whoever has most recently finished a larvae spawn, which is why the actual time of measurement matters so much, because whoever most recently spawned is *always* in the lead. To be fair, it takes quite a bit of work to calculate the effects of the spawn time rotation, but it can't simply be ignored.
Also, 18hatch is quite probably late enough that the hatch-first second queen beats the 18 hatch queen by a significant amount, increasing the amount of time during the spawn cycle that hatch-first is ahead.
Disclaimer: I've tested 14pool first and 16pool first builds vs various flavors of 15hatch, but I haven't specifically tested 11overpool/18hatch vs 14hatch/14pool, which would probably be the most reasonable comparison. The super-fast pool might be enough to keep overpool in a larva lead, but it's clearly at the cost of the mineral lead.
I wonder if double trick overpool is teh ultimate!
It's not. 12 pool off a 9OL start comes faster (slightly) than a 12overpool, and the 9OL start is more economic. The only reason you should ever double-trick is to build a pool or hatch exactly on 12, before building an overlord, but that wastes a *lot* of larvae spawn time.
Also, 10overpool is never the best option. It sacrifices too much spawn time to be useful as an economic start, and as a fast pool start, you'd be much better off 11 or 12 pooling off extractor tricks, building the pool before the overlord.
|
|
|
|
|
|
EPT
