EPT[G] Worker Transfers - Page 3
| Forum Index > StarCraft 2 Strategy |
|
Xanatoss
Germany539 Posts
| ||
|
michaelhasanalias
Korea (South)1231 Posts
Allow me to offer a quantitative rebuttal: On a map with corner mineral placement (a mineral line wrapping around two edges of your base), you'll have four close mineral patches and four far ones. In maps where the minerals are in a line, you'll have two or three close patches (Shakuras is the only line mineral setup I can think of with three patches). A close patch mines at 45/minute for single or doubled workers (12 for triple), and a far patch mines at 39/minute for single or doubled workers (24 for triple). For simplicity's sake, let's assume the most optimal situation conducive to drone transfer, which would be four close patches at both the main and natural. The issue here is that if you transfer half of your workers, you take no consideration for saturation of your main. Let me give a few examples:
HOWEVER, and this is important, if you had only transferred those four tripled workers from your main (leaving 16), you would be able to individually rally them to each close patch at the natural (I do this all the time and you can see pro gamers like idra and july doing this as well). This means you would have 12 workers on close patches and 8 workers on far patches, giving you a resultant income of 852/min. Not only is this better, but you didn't lose the travel time. As you note, you must factor in travel time, because for some time T, the workers you transferred will be providing you less overall minerals than you would have had keeping them at your base. I question your 25 second transfer time and would instead go with something closer to 15 or 18 seconds (I tested crossfire (18), shattered(17) and shakuras(15) while writing this). Now, in order to break even from worker transfer, you need to solve the following equation: ![[image loading]](http://i.imgur.com/RLiB9.gif) I(i) = Income (initial) I(f) = Income (final) t = time to break even d = distance travel time (range is generally ~15-18 seconds) * make sure you convert everything to either minutes or seconds So, let's solve t for some possible scenarios (all suppose 15 second travel time):
What does all this mean? Transferring half your workers is a bad general rule. Transferring all but 16 is easily the best general, if over-simplified rule. What you want to do is maximize the number of workers at higher income (obviously). This means you need to count the number of close patches at your natural, and then transfer that many workers at minimum. If you still have over 16 workers after transferring those 2 to 4 workers, transfer as many as you have until you reach 16. If you honestly feel like you have the spare APM at that point in time (and you shouldn't), transfer 2 to 4 workers (one for each close patch) then transfer 2 to 4 additional workers 2 seconds later and double them on each patch. Through all this, it's important to note that although you want to maximize your income, you also should factor in things like mining your main out too quickly while being unable to secure a third. In such a situation, it might be better to take the economic hit just to ensure you mine out your minerals more evenly (although this isn't the point of your write-up or my rebuttal). TL;DR - As a general rule, transfer all but 16 unless you have the spare APM to sit there and determine (then execute) what might be more efficient. If you opened hatch first as zerg, you should transfer drones equal to the number of close mineral patches at the natural and rally them to each close patch individually as you do. edit: added some of this to the liquipedia page on mining minerals because I didn't realize how skimpy it was. I really thought this was pretty cut and dry and didn't realize there were still such divergent opinions on something that always has an optimal solution (harassment risk notwithstanding). http://wiki.teamliquid.net/starcraft2/index.php?title=Mining_Minerals&stable=0&shownotice=1&fromsection=Transferring_Workers#Transferring_Workers | ||
|
michaelhasanalias
Korea (South)1231 Posts
On September 19 2011 09:33 Dulcimer wrote: I believe perhaps I should have clarified method 2 more, but the main diference between #1 and #2 isn't just their rally points, but also that in method #2 you leave 16 mineral mining workers. It is better to have an equal (and lower) number of mineral mining workers on both bases after a transfer instead of having more mineral mining workers on one base than the other base. I am going to edit the post to include some considerations with gas and how it cooperates with various builds. edit: added in the section on transferring with consideration to gas mining. You don't need to give any consideration to gas. You always transfer workers based on mineral income. | ||
|
galzohar
Israel100 Posts
| ||
|
Dulcimer
United States21 Posts
If you are FEing I agree you shouldn't follow #1 (which I said in the FE section). If however, you have 24 or more workers mining upon expanding (quite possibly less, I am going to run more tests on that) then it is best to use method 1. I'm going to do more tests and post the results (feel free to do your own tests as well) | ||
|
iSTime
1579 Posts
On September 19 2011 10:29 galzohar wrote: If they're so close, isn't transferring 1/2 more APM-efficient? Technically, how do you guys transfer efficiently (with whatever method)? If you are doing the exact same build order over and over you will know exactly how many probes you have on minerals when your nexus finishes, so you can transfer all but 16 every time. | ||
|
Barrin
United States5002 Posts
| ||
|
michaelhasanalias
Korea (South)1231 Posts
On September 19 2011 10:58 Dulcimer wrote: You make some good points Michaelhasanalias in your first post, however, generally past 24 you won't see people microing their workers, even at a pro level. As for my math, I posted absolutely no complex math. It's very easy to check. As for your math, your syntax is highly debatable. You do make a good point that it is map specific, however, I believe I already said that. If you are FEing I agree you shouldn't follow #1 (which I said in the FE section). If however, you have 24 or more workers mining upon expanding (quite possibly less, I am going to run more tests on that) then it is best to use method 1. I am going to post some replays later of various tests. The problem with your math is that you round, and then make specific deductions based off that. This is a cardinal error in mathematics (rounding or truncation error) and you need to make sure that you never let this happen when doing analysis like this where you round inputs. Only round answers, never inputs.
This leads you to some invalid deductions: Actually I'm just going to quote this passage because there are too many inaccuracies, both logical and mathematical: A base with 24 workers will bring about 800 minerals per minute, 33 minerals per worker. There are also some multibase numbers that I’ll talk about later. ..... If we follow case #2 we will lose 200 minerals of mining time over the course of 25 seconds. However, once those probes have transferred, we are mining 960 minerals per minute (8 workers on the natural, 16 in the main) as opposed to 800 minerals per minute. It will take 1.25 minutes after the transfer to make up for the transfer. If we build 24 workers and then follow case #1 we will lose 300 minerals of mining time over the course of 25 seconds (12 minerals per second). However, 2 bases operating with 12 workers each will receive an average of 1020 minerals per minute. It will take 1.36 minutes to make up for lost mining time (.11 minutes [6 seconds] longer recovery than case #2). However, once we have recovered, we receive a greater income (we receive 60 more minerals in case #1 than in case #2). A base with 24 workers will net exactly 816 minerals per minute. 3 workers per patch (except in extreme circumstances such as a few 4v4 maps with awful mineral placement) fully saturate a patch, which yields exactly 102 minerals per minute. In case #2, you transfer 8 workers, leaving 16. The mining rates of those workers from before would be 12/minute for each close patch and 24/minute for each close patch. The new mining rates would be 45/minute for each close patch and 39/minute for each far patch. The travel time on most 1v1 maps from main to natural is about 15 seconds, not 25. Just for the sake of argument, I tested three, and got 15, 17, 18 (shakuras, shattered, crossfire), crossfire being the longest I could think of. The old mining rate is 816/minute and the new mining rate would be 1008/minute (4 close 4 far) and 984/minute (2 close 6 far). You lose .3 minerals per second per worker, or 2.4 minerals per second per 8 workers. So even at 25 seconds, you're losing only 60 minerals, not 200, by following case #2 (transfer N-16). At 15 seconds, you're losing 36 minerals, extremely far from the 200 you mention. It takes only between 20-40 seconds for each of the workers to individually break even, depending from which to which patch they are coming from and going to. Following case #1, the workers are mining between 12 and 39 minerals per minute. Some of the workers you transfer (four, in fact) will never recover the minerals lost due to transfer time (about 10 minerals). The new mining rates are ~1008/min (assuming 50% probability of close or far patch doubling since you can't realistically micro at this point) for 4 close/4 far, and 984/min for 2 close/6 far. Note that these numbers are exactly the same as in case #1, except there is one notable difference: you permanently lost mining time on every worker mining at 39/min (four), or about 40 minerals. This is an immediate sunk cost realized in full over the travel time and minerals you will only get back if you mine out both the main and natural (30 minutes into the game). You claim that loss is 300 minerals, but even at 25 second travel time the loss is only 125 minerals (8 at .3/sec plus 4 at .65/sec). To realize the loss you mention, you would have to run your workers to your opponent's natural and back. Again, the economic recovery time is only about 30 seconds, not 80 seconds, with the lone difference being you have forever lost minerals per worker you transferred that ends up mining at the same rate. The only time you would ever consider maximizing the mining-out time for main and natural is when you don't plan to or can't take a third (or more) base(s). You rely on these critical errors in math and reasoning to conclude that it's better to transfer half your workers. Whenever you use quantitative analysis to qualify a position, there will be people who take you at your word, regardless of the accuracy of those numbers. Some people aren't good at math. As a person forwarding this quantitative analysis, the onus is on you to ensure the integrity of the data you're gathering and the conclusions you're making based off it. Please consider using more accurate data and adjusting your conclusion based on your new analysis. tl;dr - your math leads you to inaccurate deductions when in fact case #2 (transferring all but 16) is generally the best (albeit oversimplified) rule to follow. | ||
|
Dulcimer
United States21 Posts
@michaelhasanalias I might have not made it clear enough that that was example/theory with the 25 seconds giving an extreme case of recovery. As for your comments on my constants, I must admit I have not researched my mining constants as much as you. However, the numbers of income for various probe differences on bases is taken from my in-game-tests (which I suggest you do as well). I am going to experiment with both in actual games and if I remember analyze and post the results | ||
|
michaelhasanalias
Korea (South)1231 Posts
On September 19 2011 11:52 Dulcimer wrote: Okay, so after a large amount of testing I am getting 50-50 results (half the time being in favor of #1 and half the time being in favor of #2). After going through a large amount of the replays, it seems that #1 has more potential, and #2 is easier to use. If you use #1 and probe micro a few times (or just get lucky) then you make back your money fairly quickly. If you use #1 and you don't probe micro and get unlucky then you will still *eventually* make more money (if not due to anything but due to mining preservation of main), however it will take a much longer time to pay off. If you use #2 you will get a pretty consistent result each time, but it doesn't have as much absolute potential as #1. The original test was to measure total income (and I did probe micro in both cases because I'd get bored during the test). The second tests I did what Time did and didn't micro the probes (although I excluded the addition of gas to make it as simple as possible). I'm sorry but this is false. This isn't a question of theorycrafting, it's a question of mathematics and optimization. By definition case #1 can only ever equal case #2, and never exceed it. There are no realistic circumstances blindly transferring half your workers to your natural will result in more benefit than transferring all but 16, equal worker micro withstanding. (There are, however, situations where leaving fewer than 16 workers is optimal, but the workers transferred doesn't approach half. For example, if you have 16 workers and a new base, you could transfer four of 16 workers, breaking #2 rule, and having more optimal income. In theory, if you have 16 workers and an expansion, transferring half with perfect micro would result in maximum income, but in practice this isn't possible or applicable to any real game situation.) On September 19 2011 07:21 Dulcimer wrote: Why I don't suggest #2 The reason why #1 is more efficient than #2 is because of an effect generally called worker 'bouncing'. If 2 workers are mining the same patch, there is a slight offset in mining times. As they both mine this offset grows (due to varying distances from the mineral patch to the base). a very over exaggerated example: Suppose it takes 4 seconds for a probe to collect minerals. The travel time from the patch to the base and back to the patch is 6 seconds. There is a 1 second offset that causes the probe mining time difference to become greater over time until both probes accidentally try to mine at the same time. Eventually one of the probes will mine that patch, and the other probe will go to a different mineral patch. This causes you to lose a small bit of mining time. The less workers you have, the less times it occurs. The more workers you have the more times this occurs. Because of this, having a base with 12 workers and another base with 12 workers is better than having a base with 16 workers and another with 8 workers. Notes I have tested each case mentioned in this post in game to make sure the numbers align correctly. Zerg The issue of transfering becomes much more complex with zerg. As a zerg player you will expand sooner (often times the expansion will be considered a fe of sorts) and you have the ability to make multiple workers at once (which will change the way you want to transfer a bit). I suggest reading the Fast expansion section of the conclusion which is the closest I come to a definitive answer for zerg. I might do another post on zerg transferring and droning sometime if I get the chance and some thoughts of interest occur to me. Just a few extra notes on flaws in reasoning: Each worker already mining at the same rate he would mine when he is transferred would need to bounce at least 3 times to make it worth transferring. Bouncing only really occurs once you get more than 16 workers (or have between 9 and 16). But another problem is that you'd have to transfer before the bouncing occurs. If you let them bounce around before transferring, and then pull off all but 16, they won't bounce anymore once they stabilize (a maximum of 1 or 2 bounces each, usually less). Zerg also is much simpler, not more complex. Without consideration to risk of harassment, when you hatch first, you transfer drones equal to the number of close patches you have. In the case where you may try to cheese your opponent with only a finite number of drones, you can transfer double that amount. Here's a great video of july zerg doing exactly this in the GSL when he went for a hatch first 6:00 baneling bust off just 16(15) drones. http://www.gomtv.net/2011gslsponsors2/vod/61321 (set 5 on metalopolis) Also, with zerg, ideally you only want 2 per patch anyway because you will be expanding more often, so rule #2 simply doesn't apply. However, under no circumstances would you ever just transfer half your drones. Instead, you hatch a new round of drones and rally them to the new bases, or from main to nat what I mention above. | ||
|
Dulcimer
United States21 Posts
As it is, I am getting tired and you have more evidence. So I will edit the post to contain your opinion and run more tests some other time and make sure that all the evidence still holds true and what have you. Thank you for your contribution Ah now you have edited your post after I posted : P Bouncing seems to happen more to me, but I may very well be mistaken, I will do more tests on that at a later point (or look for more research on it). Zerg hatchery droning is very complex because of early pressure generally it seems to me. | ||
|
KimJongChill
United States6429 Posts
I've always transferred half of my workers after a 14h, so does this mean that this might not be so good anymore against a terran doing 2rax pressure? | ||
|
Tortious_Tortoise
United States944 Posts
| ||
|
HaRuHi
1220 Posts
Maybe bouncing is a much bigger factor in it than I realized yet, so cloning with spare apm might be the answer. Other than that I sometimes send half just because I need them at my natural to help defend.  | ||
|
arbitrageur
Australia1202 Posts
On September 19 2011 07:31 Dulcimer wrote: No, in short term it is better to not transfer any workers Do you possess evidence that quicker use of the 18th-~25th larvae outweighs an increase in the marginal efficiency of the workers transferred? | ||
|
arbitrageur
Australia1202 Posts
On September 19 2011 13:03 HaRuHi wrote: In Theory, leaving 16 is optimal. You have not provided any basis for this claim. You haven't accounted for the fact that transferring leads to a short-term income reduction, which delays the transformation of the 18th-23rd (maybe more) larvae into drones. | ||
|
jimbob615
Uruguay455 Posts
you often need at least 6 SCVs at your natural to repair bunkers in case protoss or zerg do a push up your ramp | ||
|
askTeivospy
1525 Posts
| ||
|
Crushinator
Netherlands2138 Posts
I would consider not transfering half as zerg because zerg seems to have the highest time preference for minerals. More minerals early in the game may mean more drones. | ||
|
Fairwell
Austria195 Posts
On September 19 2011 11:36 iSTime wrote: If you are doing the exact same build order over and over you will know exactly how many probes you have on minerals when your nexus finishes, so you can transfer all but 16 every time. True, but even if you don't or you maybe lost some workers due to harassment etc earlier before your expo is up and running, just ctrl+leftclick on your workerline the main or box them, if you have both gases taken by then all workersthat are more than 2 and a half line are too much and need to be transfered (2 less if you only took one gas, again 2 less if you didn't take any gases, since you will only see two workers for each gas). So with some practice you will just box quickly, see lets say 1 less than 3 rows (and knowing you have 2 gas), you select 3 workers and send them over knowing you have the right amount of workers. I really recommend trying this out, it's a very efficient method in my opinion. I've been doing this for a long while already. | ||
| ||
