|
While practicing some builds in SP and trying to fine tune the early timing, I kept noticing that sometimes I could afford things when I wanted to, and sometimes not. After doing a bit of forum searching, I couldn't find any hard numbers about the effects of early worker micro on econ, (plenty of vague conjecture though) so I did some nerd science. Here are the results:
Experiment:
Determine the relational benefit, if any, of manual worker pairing in the first few minutes of the game.
Points of control:
To make sure that only the manual pairing was being tested here, I used a few techniques to isolate this process.
1. All tests were done on the YABOT version of Shakuras Plateau, in the top right spawn point. The initial first actions were done one time, with no worker split. (this is not a test about initial worker splitting) This was then saved in single player, and all tests were loaded from this point. The reason for this is to prevent initial lag/mechanics on my part from affecting the results.
2. The 8th probe was rallied away from the nexus and is relegated to pylon production. This is to prevent the interaction of pulling a worker to build/return from affecting the results. This would be similar econ to a forge expand build in which the probe builds the pylon and then scouts.
3. No gas was made, for similar controlled circumstances as #2. The choice of which workers to pull might be negligible but there's no reason to risk it.
4. All tests end at the 5 minute mark. Keep in mind that the economic effects may be realized well before then. It was just an arbitrary stopping point.
5. Tests were done with varied numbers of total probes built, some at 17, some at 23, and a couple at 25. Each run also produced 2 pylons. The probes and pylons built were accounted for in determining the total amount mined in 5:00.
The first runs in the test were just to see how efficient the mining AI is by itself, stopping at various worker counts. The rally point was simply set at the closest middle patch and left there. The resulting total mining actually varied up to about 25 minerals of difference, which suggests that there is a degree of randomness in the AI. Sometimes the AI finds an equilibrium just after the last worker is built, sometimes one of them wanders around like a jackass and doesn't accomplish anything for awhile.
17 workers then stop (16 mining) no micro - no chrono - averaged over 5 runs
average : 2431 mined at 5:00
17 workers then stop (16 mining) no micro - x3 chrono - averaged over 5 runs
average: 2521 mined at 5:00
19 workers then stop (18 mining) no micro - no chrono - averaged over 5 runs
average : 2503 mined at 5:00
23 workers then stop (22 mining) no micro - no chrono - averaged over 5 runs
average: 2522 mined at 5:00
The next tests involved 'rally pairing' each new probe as it is constructed to an appropriate patch, thus committing it to a patch with no wandering. This was done at normal speed (faster) and is not difficult to do. Unless you're zerg, lol. note: it is possible to make 'trios' for the further away patches, but the AI will not allow you to do this until every other patch is being mined at the moment it makes the decision.
This did not vary much, so no average is needed for any further test. +- 5 minerals is the expected error rate.
17 workers then stop (16 mining) rally pairing - no chrono
2485 mined at 5:00
17 workers then stop ( 16 mining) rally pairing - 3x chrono
2580 mined at 5:00
Next, the proverbial upper limit of worker micro was tested. You may have seen pros doing this recently, redirecting the early probes such that the 3 closest patches are paired right away. Further probes were rallied to the middle patches, etc. I did this test on the 'slower' time setting to be sure that it was optimal. Gosu players could probably do this on faster but my stupid fingers can't. WRU robot devil.
ex pic of fast pairing: (my screenshots aren't working, no idea why... mspaint drawing instead GO): + Show Spoiler +
17 workers then stop (16 mining) fast pairing - no chrono
2520 mined at 5:00
17 workers then stop (16 mining) fast pairing - 3x chrono
2610 mined at 5:00
19 workers then stop (16 mining) fast pairing - 3x chrono
2695 mined at 5:00
25 workers constant production till 5:00 (24 mining) fast pairing UP TO 16 - 4x chrono
2750 mined at 5:00
25 workers constant production till 5:00 (24 mining) full fast pairing - 4x chrono
2800 mined at 5:00
Conclusions/notes
So, from the testing, it seems that the following conclusions can be drawn:
1. AI mineral finding is a little randomized. No micro on your workers will result in inconsistent early timings.
2. Rally pairing manually produces about 55-60 more minerals with your first 16 workers. Fast pairing (if executed flawlessly) produces about 90 more minerals with your first 16 workers.
note: While the technical timing for this econ boost is at 5:00, most of the bonus comes by about 2:45-3:15 depending on race, which is when the manual pairing stops. Occasionally the non micro'd workers will waste another 5-15 minerals by pathing badly over time.
3. This process faces heavy diminishing returns after 16 workers. While there is still some more econ to be squeezed out of it, at this point you are going to be doing more active scouting and will not have time. In the case of protoss, with 3 chrono boosts on probes, this occurs at 2:40.
It's up to you to decide whether having this money at ~3:00 is worth the trouble. As a point of perspective, fast-pairing workers produces almost exactly as much extra econ as three full chrono boosts at earliest opportunity! I would argue that this is more useful than spamming 1234, at any rate.
I used the load function during my tests, which in retrospect was not the best idea because now I don't have any replays saved. edit- I made an example run of each type here
|
Very nice post. I've been microing my initial drones (yes, im zerg, its a pain to time sometimes  ) for a long time now, mostly because I have nothing else to do in the early game, but I gotta be honest I have definately felt the difference between doing it and not doing it. I feel like I can get my spawning pool up just those few seconds faster, and over a lot of games where this happens consistently, its a huge deal. But its great to actually see some hard numbers. Those ~50 additional minerals this produces is one free drone that you get to keep for the entire game.
To put things in perspective; assuming you end up around 70 workers (= ~56 mining minerals) this is basically 1,8% more income in the end game if you micro than if you dont. And its 6,6% more income around the time you have 16 workers. Big deal? Yes!
Great job getting the numbers done 
|
genius, love the stats. I've always been curious about the exact effects and how much more minerals you could get from it.
One cool thing that would be awesome is maybe a guide or a replay of how to perfectly micro your workers. I've tried, messed up and given up xD
but good post
|
Dominican Republic913 Posts
|
Really good post!
|
With really good mineral pairing, a few timings I know as Protoss are that you can get your 9 pylon as early as 46 seconds, a 13 gate at 1:35, and a subsequent 14 gas at 1:57 and you're able to drop your cyber core at ~2:40. Basically gives you a 10 second advantage over normal mineral mining, which makes a huge difference in PvP.
Nice find by the way, good to see some numbers!
|
I've been working through this by myself both in-game and in my head for months, and as I improved at basic pairing I definitely noticed my timings got much easier to hit. Nice to see some tangible numbers to reinforce this.
|
Thanks for the numbers I've been waiting for someone to actually calculate this, but for your second trial do you mean that you microed where the workers spawned to hit unused patches or grabbed each spawning worker and locked them on a patch? or both?
|
I don't think pairing is necessarily "more economic" (like your first 6 workers) but obviously making sure no time is wasted on path finding is going to benefit your well timed pushes.
Good post I don't know strategically what this has to do with anything, as every little itty bitty advantage you have before the 5:00 mark is lost after the first battle. But I guess if you are going for a timing push and you can get an extra unit out that is a bonus.
|
On June 18 2011 15:10 Navillus wrote:
Thanks for the numbers I've been waiting for someone to actually calculate this, but for your second trial do you mean that you microed where the workers spawned to hit unused patches or grabbed each spawning worker and locked them on a patch? or both?
I've included an example replay of each type at the bottom of my original post. By 'rally' pairing, I mean that (with a little practice) I was able to anticipate a pair being naturally formed by my rally point based on the timing of the spawn. If done correctly, this will not normally need any manual selection of the probes. (I'm not sure that explanation actually helped lol)
If you want to see good examples of manual pairing in action, check out LiquidTyler's games during MLG Columbus. He is fairly good at this technique! There may be more that do it, but I specifically noticed this today from Tyler.
|
great post.... but out of curiosity (not directed negatively at the OP by any means) - how come all these great maths and scientific calculations of sc2 mechanics come from people with low post counts? I mean, there are obviously lots of great posts from people with high post counts too.... it just seems like when it comes to really crunching serious numbers and doing hundreds of tests in perfect environments, it's usually the people with low post counts...
or maybe i'm just imagining things
either way, GREAT post!
|
Awesome findings. I'll start doing this just because now, it was never really clear whether it had a significant effect or not.
But 50 minerals is indeed an extra drone or extra zergling in case of early pressure.
|
Finally someone sat down with the scientific method and got some numbers for us lazy people. Well done.
Edit: I'm not sure this is really a discussion, more of a guide or something.
|
In the scientific method we trust; thank you for your data, hard work, and number crunching.
Data leads me to believe that I should be doing this manual pairing thing.
|
On June 19 2011 13:55 Soluhwin wrote:
Finally someone sat down with the scientific method and got some numbers for us lazy people. Well done.
Edit: I'm not sure this is really a discussion, more of a guide or something.
Not even really a discussion or a guide... just a confirmation that we should all be working on mineral pairing
|
Tight man that could be an extra zealot!
|
On June 18 2011 15:49 Kornholi0 wrote:
I don't think pairing is necessarily "more economic" (like your first 6 workers) but obviously making sure no time is wasted on path finding is going to benefit your well timed pushes.
Good post I don't know strategically what this has to do with anything, as every little itty bitty advantage you have before the 5:00 mark is lost after the first battle.
this type of drone micro play is the standard of zerg openings, if you are going to 15hatch you need it down asap, a few seconds earlier might be the difference between being ready for a bunker rush or not. Those few seconds come from having the fastest mineral generation possible in the first few seconds, like, with your first workers. I don't look at it as an extra worker or unit, I look at it as not dying to early all-ins, as the sooner creep is at my natural, the sooner I am safe from instant loss to proxy buildings.
Great post with hard evidence behind it. This is the sort of thing that can improve your play at any level if you don't currently do it.
|
Apologies if I am mistaken,but this topic is about making sure workers dont have to travel to find an empty mineral patch, correct?
If that is the case then yes, I also try to manually do it from rally point to the empty node, but also I try to do it to a mineral patch where the previous worker just finishes gathering.
|
This is very important and very awesome i see idra doing this all the time on his stream
|
To do a true scientific test you'd be better off doing something akin to the worker micro in skrag's build order test map and using the fixed random seed option in the map editor. I trust AI in game results much more than human trials as I can download the map and repeat the tests myself.
|
|
|
|
|
|
EPT![[image loading]](http://i.imgur.com/JkoRv.jpg)
