EPTI'd rather just guesstimate for my own conclusions, since any sort of fixed number you try to give is just going to be a very rough approximation anyway, and provide the graphs and raw data so that people can draw their own conclusions if they want.
Pushing The Limits of Zerg Economy Builds - Page 19
| Forum Index > StarCraft 2 Strategy |
|
Skrag
United States643 Posts
I'd rather just guesstimate for my own conclusions, since any sort of fixed number you try to give is just going to be a very rough approximation anyway, and provide the graphs and raw data so that people can draw their own conclusions if they want. | ||
|
Skrag
United States643 Posts
On December 26 2010 00:39 Nolari wrote: Wow Skrag, awesome sauce! Now it's not just an investigation into a couple of builds, but a framework to investigate any conceivable build with fairly little effort. Well, it's still a fair amount of work to add builds, because I do have to go through and get the right timings, figure out which workers to pull to try to build stuff optimally, etc, which still involves a decent amount of trial and error. But that's *way* easier than getting a few timings, adjusting drone micro, getting more timings, adjusting the drone micro, etc.  | ||
|
Skrag
United States643 Posts
8 seconds of hatchery time is worth .533 larva, 4 seconds of queen time is worth .4, for a total of .933. | ||
|
Nolari
Netherlands51 Posts
On December 26 2010 00:48 Hurkyl wrote: (2 minutes because that's an integer multiple of both 15 and 40 seconds) Energy regenerates at a rate of 9/16 energy per game seconds, so injects happen every 25 / (9/16) = 400 / 9 = 44.444... game seconds, not every 40 game seconds. | ||
|
Hurkyl
304 Posts
On December 26 2010 01:20 Skrag wrote: The problem with trying to get averages is that the result depends purely on what you decide to calculate. A 2 minute average is going to be different than a 3 minute average, etc, and the reality of the way larvae work means that there are clear periods where the differences can be large. And 2 minutes is a *really* long time game-wise anyway. I'd rather just guesstimate for my own conclusions, since any sort of fixed number you try to give is just going to be a very rough approximation anyway, and provide the graphs and raw data so that people can draw their own conclusions if they want. A numerically computed average is surely going to be better than one formed by eyeballing the data.  Larvae production is eventually periodic, and so averaging over one period captures the most significant statistic. You could always average over a sliding window if you're worried about it -- e.g. at time T, plot the larvae average over the interval [T, T+W) for some suitable window width W. Or maybe convolve the data with some other smoothing function. | ||
|
Hurkyl
304 Posts
On December 26 2010 01:32 Skrag wrote: How do you get 1 and a third though? 8 seconds of hatchery time is worth .533 larva, 4 seconds of queen time is worth .4, for a total of .933. I hadn't noticed that you weren't starting a Queen immediately upon Hatchery completion. | ||
|
Skrag
United States643 Posts
| ||
|
Skrag
United States643 Posts
On December 26 2010 02:26 Hurkyl wrote: I hadn't noticed that you weren't starting a Queen immediately upon Hatchery completion. Oh, yeah, I'll have to go look at why that is. Pretty sure I've tried to get queens started ASAP though, so there's probably a good reason, as opposed to just being sloppy. | ||
|
Skrag
United States643 Posts
| ||
|
Skrag
United States643 Posts
I suck at absolute vs relative spreadsheet references, and copied the wrong cell when filling in some data. Deleting the post so I don't confuse anybody that isn't already confused. :/ | ||
|
Skrag
United States643 Posts
*EDIT* I just republished the map after adding a few more builds, so this replay may stop working. 13h15p Replay If somebody wouldn't mind trying it out and letting me know if it works, I'd appreciate it. | ||
|
ChickenLips
2912 Posts
On December 26 2010 04:23 Skrag wrote: One more shot at sharing replays. I switched things around so that the map is playable on battle.net, and let this game run *on* battle.net rather than in the map editor. 13h15p Replay If somebody wouldn't mind trying it out and letting me know if it works, I'd appreciate it. works flawlessly | ||
|
Skrag
United States643 Posts
| ||
|
Skrag
United States643 Posts
16h15p (purple) 13h15p (brown) 14p16h (light blue) 11p18h (red) 15p16h (orange) 13p15h (dark blue) ![[image loading]](http://www.mediafire.com/imgbnc.php/de637bda66404c14a1e81e5862118a8b67996edf2df6c8b38f1b7eeb3490fbdd6g.jpg) The results of this latest round of tests are slightly different than the last round. I changed the map so that I could publish it on battle.net and let the game run while actually playing there, so that replays could be posted. One thing that shows is that very simple random differences can have a pretty significant impact, so the absolute numbers can't really be taken as gospel. For example, before reconfiguring the map, 11p18h was only about 95 resources behind 14p15h, but now it's closer to 130. The larvae graph unfortunately got really difficult to read as I added builds to it. The easiest way to see what's going on there is going to be to download the spreadsheet (you'll need openoffice, which is a free download), and modify the graph to just show one at a time. Imprecise commentary on the larva graph: 16h15p is slightly behind the entire time, while 13h15p is slightly ahead. The other 4 builds are all slightly behind until a certain point, and then pull slightly ahead in larva count. 11p18h catches up at about 4:00 13p15h catches up at about 4:15 14p16h catches up at about 4:20, and ends up further ahead than any other builds 15p16h catches up at about 4:30 Here's a graph showing the 2 minute average of the larva difference. Obviously it's really difficult to be precise when there are instantaneous jumps of 4 at a time, but this seems to give a decent picture of which builds are ahead in what approximate timeframe. The catch-up times I listed above are fairly accurate, but the average gives a way of displaying all the builds on a single graph without a ton of overlap. This graph would seem to indicate that 14p16h is ahead of 13p15h even though I said 13p15h was the larvae leader. In that respect, I'd be more inclined to believe the average graph, since I was just eyeballing and guessing. ![[image loading]](http://www.mediafire.com/imgbnc.php/f1c24a3074f6f4054cc6637a7861131ce7fe59a8b968c8d2d7c0966c9da282006g.jpg) Commentary on each of the builds measured: POOL-FIRST BUILDS: 13p15h: despite being economically ahead of all the other pool-first builds in the drone race, I honestly don't see any reason to ever use this build at all. It does have a pretty significant larva advantage over most other builds measured, but it never has enough resources, and gets stuck unable to use larvae a *lot*. 11p18h: This build held up to the reality injection very well, better than I expected, and probably far better than any of the people claiming that it would fall apart economically once you started trying to actually do anything would have thought possible. 14p16h: Somewhat unsurprisingly, since a 14pool was commonly quoted as a build that would be beating 11p economically, this one held up pretty well. It has a 20-50 resource advantage on 11p early, and ends up in pretty much the same spot as 11p after max saturation, being around 130 resources behind 11p. This build also appears the be the larva leader, and was actually able to take advantage of the extra larvae, so that might be a consideration. 15p16h: This build plays out pretty similar to 14p16h, except that it falls further behind as max saturation is approached. A big part of that is that this build had a *really* hard time matching the timings I was going for (especially the spine crawler), so this might be a little bit unfair. But without doing testing under different circumstances (which I probably won't have the patience to do), it seems there won't really ever be a reason to use 15p instead of 14. I'm pretty sure 14p is far more common, and the problems 15p had in executing the things I wanted to execute probably explains that quite a bit. HATCH-FIRST BUILDS: 13h15p: This build gains a slight larva advantage over 14h15p (somewhere in the neighborhood of one larva) at the cost of lagging behind in resources by 30-40. It *might* be less susceptible to hatch blocking, although the hatch times are close enough that I'd be surprised if there was really that much of a difference. 16h15p: This build trades a small larva disadvantage for a small early resource advantage. Where 13p15h didn't have enough resources to use it's available larvae, this one doesn't seem to have enough larvae to spend its resources. If you were planning a very early 3rd hatchery, or expected to need to build a bunch of spine crawlers, this one might do ok, but you'd have to be planning on doing that in the first 5 minutes for it to be worthwhile, because the early advantage is only 25-ish resources, but the later disadvantage is 65-ish. 15h14p: The data for this build is not posted in the graphs, but it appears to take a late larvae advantage over 14h15p, at the cost of a slight early larva disadvantage and a resource disadvantage that is larger than 13h15p. So if you're wanting to trade resources for larva, it seems 13h15p would be the better choice. | ||
|
alepov
Netherlands1132 Posts
| ||
|
Minus`
United States174 Posts
On December 27 2010 06:50 Skrag wrote: Graph of the resource difference between 14h15p and the following 6 builds: + Show Spoiler [Graph] + 16h15p (purple) 13h15p (brown) 14p16h (light blue) 11p18h (red) 15p16h (orange) 13p15h (dark blue) I edited 13h15p into the script a while ago and showed similar results to yours in an earlier post (lost by 160 minerals at 9 minutes for me, which was behind the 11/18 build). But, from the stats your replay generated, I had 13h15p beating 14h15p -- and any other build -- in resources at every point after 5:50. Did I manage to do something wrong here, or have I misread the graph/results or something? (BTW, great work Skrag, you're the man.) | ||
|
TheUberMango
United States77 Posts
| ||
|
Skrag
United States643 Posts
On December 27 2010 07:07 alepov wrote: so as it stands now, 14h15p is the "best" ? Was that ever a question? The economic tests showed 14h15p and 15h14p being *extremely* close economically (which makes perfect sense, because the only real difference is a tradeoff of a few seconds between hatch and pool times). I might add 15h14p just out of curiousity, but mainly I wanted to compare vs an earlier and a later hatch. | ||
|
Skrag
United States643 Posts
On December 27 2010 07:21 MinusPlus wrote: I edited 13h15p into the script a while ago and showed similar results to yours in an earlier post (lost by 160 minerals at 9 minutes for me, which was behind the 11/18 build). But, from the stats your replay generated, I had 13h15p beating 14h15p -- and any other build -- in resources at every point after 5:50. Did I manage to do something wrong here, or have I misread the graph/results or something? (BTW, great work Skrag, you're the man.) The readjustments I had to do in order to make it playable on battle.net improved the performance of 14h15p by 15-20 minerals for some reason. In the graph I just posted, 14h15p's resource count under the new configuration (apparently forcing a fixed random seed via trigger ends up using a different fixed seed than forcing a fixed seed in the map editor, so I had to rearrange some stuff to make them work properly again) was used as the baseline, so what's actually graphed for each build is its resource count minus 14h15p's resource count. So if you were comparing the replay I posted against old data, it could be quite a bit closer, and 13h15p might even seem slightly ahead. But I also ran 13h15p before I had to adjust everything, and the results were practically identical, a slight larva advantage and a slight resource disadvantage. Also, if you put 13h15p into the script and ever showed it being behind 11p18h at the 9 minute mark, then I would have to say the script you added was extremely sub-optimal, because every incarnation of 13h15p I've done was significantly ahead of 11p, but slightly behind 14h. | ||
|
Skrag
United States643 Posts
The old links should still work, here they are again: Spreadsheet (OpenOffice required) The Map Also just added a 2 minute larvae difference average graph to the earlier post. It's less precise than the instantaneous larvae measure obviously, but allows all 6 builds to be shown on the same graph. | ||
| ||
