|
hi guys, many of you have probably read the great piece Do you macro like a pro? by whatthefat: a study of stats from 2,100 games showing that pros make more workers than noobs, consistently, all through the game.
here's a breakdown of workers-per-minute (wpm) from 34,000 games, graciously shared from by drop.sc from their public replay collection:
lesson: to improve your game, you should improve your worker production.
but, how can you tell if your worker production is really improving?
i've made a site called ggtracker to solve problems just like this. to let you track your stats, coach yourself and get a proper full-on nerd handle on your self-improvement.
you get charts like this:
in this article, we study the workers-per-minute from 34,000 games, introduce the ggtracker system you can use to track your own performance, and uncovers one interesting new finding not covered by whatthefat.
check out the article, it has interactive version of these graphs and explain ggtracker a little more.
hope you like it... let me know what you think! --dj
|
Cool, constant worker production is definitely one of the major pillars of sc2 success and its not always easy to tell how well you are doing at it if you just miss a few seconds here and a few there.
|
Can we get further breakdowns instead of clumping Bronze-Diamond into one? I'm currently Diamond with an average WPM of 2.4. I imagine the data is skewed because of the mixing of lower leagues.
|
Oh my god, I'm uploading replays, and your status updates are hilarious.
Are you intending on adding extra statistics?
|
|
|
I have to quote this and talk about it...
"for the super nerds out there, a technical note. due to limitations of the replay file format, we cannot measure worker production exactly; we can only measure when the player attempts to create a worker. in some cases the player can spam worker-creation commands that do not correspond to actual worker production. therefore we put a 3-second anti-spam filter into the measurement. it’s not perfect. it’s at its worst when a player will queue up probe production or make six drones in a second. despite these problems, there’s a clear relationship between wpm as we measure it and league, so for that reason i’d argue you’re pretty safe using this wpm as a metric to measure your own improvement as a player."
sad that this isnt very accurate then 
|
|
|
Fascinating article and beautiful graphs. May I ask what package you used to create the graphs?
|
Could you incorporate supply vs time also? IF so that would be the bomb.
Here's mine lol
I think the workers per minute is a bit misleading. I'm a plat Terran with 2.6WPM average and my workers per minute game graph says i make more workers than masters and above by a considerable margin,
Other than that, I love this ALOT!
|
On March 10 2012 15:34 TheGreenMachine wrote:I have to quote this and talk about it... "for the super nerds out there, a technical note. due to limitations of the replay file format, w e cannot measure worker production exactly; we can only measure when the player attempts to create a worker. in some cases the player can spam worker-creation commands that do not correspond to actual worker production. therefore we put a 3-second anti-spam filter into the measurement. it’s not perfect. it’s at its worst when a player will queue up probe production or make six drones in a second. despite these problems, there’s a clear relationship between wpm as we measure it and league, so for that reason i’d argue you’re pretty safe using this wpm as a metric to measure your own improvement as a player." sad that this isnt very accurate then
Think of it as a reliable proxy for your worker creation rate, rather than a literal counting of your workers. If you build more workers in a useful fashion (the 3-second rule encodes "useful fashion") then the measure of WPM will increase. So it remains useful as a tool to track your worker creation habits.
|
thanks thisillusion! i'm using highcharts, a nice javascript charting package.
iTzSnypah, thanks for the suggestions. supply vs time is going to be approximate, unfortunately, but we can do it. maybe you really do make a lot of workers? you might be that rare Plat player who's really got the making-workers part of the game down cold.
|
|
|
On March 10 2012 15:48 dsjoerg wrote:thanks thisillusion! i'm using highcharts, a nice javascript charting package. iTzSnypah, thanks for the suggestions. supply vs time is going to be approximate, unfortunately, but we can do it. maybe you really do make a lot of workers? you might be that rare Plat player who's really got the making-workers part of the game down cold.
Well after further thought, supply vs time would only be good until about 15 minutes because then your usually max supply.
|
Dumping my replays on board for the lols :D
|
On March 10 2012 15:48 thisillusion wrote:Show nested quote +On March 10 2012 15:34 TheGreenMachine wrote:I have to quote this and talk about it... "for the super nerds out there, a technical note. due to limitations of the replay file format, w e cannot measure worker production exactly; we can only measure when the player attempts to create a worker. in some cases the player can spam worker-creation commands that do not correspond to actual worker production. therefore we put a 3-second anti-spam filter into the measurement. it’s not perfect. it’s at its worst when a player will queue up probe production or make six drones in a second. despite these problems, there’s a clear relationship between wpm as we measure it and league, so for that reason i’d argue you’re pretty safe using this wpm as a metric to measure your own improvement as a player." sad that this isnt very accurate then Think of it as a reliable proxy for your worker creation rate, rather than a literal counting of your workers. If you build more workers in a useful fashion (the 3-second rule encodes "useful fashion") then the measure of WPM will increase. So it remains useful as a tool to track your worker creation habits.
My experience with examining my own replays in sc2gears is that the worker production count is grossly inaccurate when looking at zerg replays. I'll frequently see 3-8 workers "produced" for each single one in the early game as I hold the D key down with larvae selected while waiting for 50 minerals.
|
When considering machine's worker production, have you considered that the peak you see in machine's is simply because he plays Z while the full dataset is made up out of all three races? The spike is visible in the 'full set' too, just less prevalent due to only a third (or something) being Z replays.
|
your Country52797 Posts
I'm quite surprised there are so few games of 3+ workers per minute, as that's not even constant 1 base probe production.
Going to give this a shot.
80 workers in a 28 minute game=2 workers per minute?
|
On March 11 2012 09:49 TehTemplar wrote:
I'm quite surprised there are so few games of 3+ workers per minute, as that's not even constant 1 base probe production.
Going to give this a shot.
80 workers in a 28 minute game=2 workers per minute?
Well you need what, 24 workers per base? With mules something like 60 as Terran. Knowing when to make drones is a very important thing to learn for a Zerg player. No reason to make drones when you're not getting a base to put them to work to, right? Knowing when to stop SCV production and how many OC's to make is a good sense to have for Terran. For protoss? Dunno, never played them. 70 Probes, that's it?
|
I made 78 workers in an 18 min game and it says my WPM is 1.6 Or a 27 minute game where I made 115 workers it says that my WPM was only 0.9? Im a zerg player BTW so maybe its the 3 second rule bogging down my stats?
Also breaking down that stats by race would be sweet. Id assume Zergs would have a higher WPM than the other races so seeing how you stack up to master zergs players would be more helpful than comparing my results to a mix of Master players of all races.
|
Whoa. There's a nice general trend from bronze to silver to gold to plat to masters, and then GM is totally different. I guess that shouldn't be too surprising, though, since there's a huge jump between low and high masters...
|
By the way, the name of your website happens to coincide with the name of a trojan for the Android phone =(
|
i know, crazy huh? i had nothing to do with that trojan i promise! pinky swear!
|
SKGZombie, thanks for the feedback. Can you send me the link to the game page and I'll investigate closer and see if it's the 3-second antispam that's responsible, or maybe I have a horrible bug I need to fix!
|
isn't sc2gears able to measure worker production exactly? It might take a while if you try to analyze 20k games or more at once but i am pretty sure that would lead to more interesting results. I also wonder if the difference in higher skill regions is just because people play more standard and therefore the game goes to a lategame scenario more often which would change the first graph but iam not really sure into which direction. Probably depends what race you are and how long the game goes on after you have enough workers.
I wonder if there is a way to exclude replays in which one player cheesed another player or had to cut a lot of workers due to an allin timing attack or some other push etc.
But yeah i don't really know if this data helps anything at all, ofc making workers constantly is a good thing but there could be lots of simple things that can change the outcome of the data.
e.g.
master players are able to manage their drops better -> therefore zerg players need to build more drones and thats why the zerg opponent needs to make more drones than a diamond zerg player with an unharrassed economy.
Or a Zerg master leaves right after he knows he can't stop some allin and therefore probe cutting for his enemey is not as decisive as it would be if the zerg kept trying 5 more minutes.
These example seems to be cheesy but i hope you get what my overall problem with this data is. I much rather would want a worker production over gametime graph for each race and weed out non standard games(games under 5-6 minutes, games in which a huge amount of workers was produced in the lategame you als have to discount drones in the lategame because they often just will be made to be turned into spinecrawlers etc.) to get more accurate data, i know that this change wouldn't turn the data into something perfect but it would help to make it better.
|
if i would give 1 advise to a player whos is not diamond/master is just make workers non stop and just spend your money even if its just 20 baracks or 10 gates you will get to diamond/master just by doing it
|
On March 11 2012 21:37 taitanik wrote:
if i would give 1 advise to a player whos is not diamond/master is just make workers non stop and just spend your money even if its just 20 baracks or 10 gates you will get to diamond/master just by doing it
Not always true
Spending money, of course,but not as blindly as making 20 barracks etc.
The current league is much more complex and difficult compared to the league last season or last year. It becomes more competitive and difficult each day. SO while your advice could work for past seasons, now it is more complex.
|
On March 10 2012 15:34 TheGreenMachine wrote:I have to quote this and talk about it... "for the super nerds out there, a technical note. due to limitations of the replay file format, w e cannot measure worker production exactly; we can only measure when the player attempts to create a worker. in some cases the player can spam worker-creation commands that do not correspond to actual worker production. therefore we put a 3-second anti-spam filter into the measurement. it’s not perfect. it’s at its worst when a player will queue up probe production or make six drones in a second. despite these problems, there’s a clear relationship between wpm as we measure it and league, so for that reason i’d argue you’re pretty safe using this wpm as a metric to measure your own improvement as a player." sad that this isnt very accurate then
indeed. as a zerg, injects pop 4 larvae at a time.. so only 1 out of 4 gets counted. Sometimes you get supply blocked as zerg and make >10 drones at a time, up to 20 sometimes.
Only one gets counted?
|
kaluroo, unfortunately what you're describing is true. i will definitely work on finding a better way.
in the meantime, is the current metric a useful one to track, for a player who's trying to improve? perhaps the metric should be called not "workers per minute" but "worker creationwaves per minute".
once properly understood, i think the metric could be useful.
put it this way -- would you expect a better player to have more creationwaves per minute or per game? across all 40k players, the # of creationwaves is clearly correlated to league. will do a per-race analysis soon as well. --dj
|
Are you counting people who are like for example 4v4 masters as masters across the board. Not to step on the toes of team game lovers but i think it should be based on 1v1, high level team games are generally pretty cheesy.
|
lksf, yes currently ggtracker shows the player's highest league across all gametypes. we'll try to make the site more clear about this in the next version!
|
Pretty sweet, although it's accusing me of becoming lazy and complacent recently.
It would be nice if the match numbers shown on the graphs were listed in the matches table too. That way I could go straight to the matches where my worker production was nice and high, and watch those replays, instead of having to browse through pages and pages looking for wpm values.
|
From what I read so far, the way the correlation between workers and level works is far from clear : actually, it's maybe much more because you are better, that you manage to make more workers; than the contrary.
|
Macpo, point taken that correlation is not causation.
netherh, great idea! Will do it in the next version.
|
On March 11 2012 21:01 idonthinksobro wrote:
isn't sc2gears able to measure worker production exactly? It might take a while if you try to analyze 20k games or more at once but i am pretty sure that would lead to more interesting results.
No. The problem is in the SC2 replay format. It is basically a list of actions taken by both players (build-order for unit X, move unit Y to location Z, change camera to location Z, etc...). The game can then replay these actions and since there is no random element in the game engine, the result is exactly the same every time. Replays also contain failed commands, such as building a worker when you dont have the minerals for it. Without the original game engine to determine exactly which build commands were successful, the best you can do is apply clever guesses to estimate how many workers were actually built. It's possible that SC2Gears has a better guess-algorithm (I wouldnt know). In that case, it'll be easier to just implement that in the script that these results were generated with.
|
On March 12 2012 04:44 Macpo wrote:
From what I read so far, the way the correlation between workers and level works is far from clear : actually, it's maybe much more because you are better, that you manage to make more workers; than the contrary.
I think part of it is that the better you are, the more likely you are to go for macro builds. If you go Reactor Hellion into 3CCs, you can build a lot more SCVS per minute, which is probably why there's a huge spike for GM players.
|
On March 10 2012 14:44 dsjoerg wrote:
why do the pros have lower wpm at the beginning of the games?
|
On March 12 2012 08:29 parkin wrote:why do the pros have lower wpm at the beginning of the games?
The x-axis is wpm and the y-axis is population density. Meaning, the average pro has just a little under 2 wpm, and the rest are sort of bell-curve-distributed around that, while the masters average lower and have a differently shaped distribution.
|
hello,
I love your website, very cool to see how my trends compare to that of the masses and to the masters and above trends
Just an idea, you should create an option to allow players to seperate their games based on length, and allow them to see the trends of those games
For example, the WPM of all 10 minute or shorter games for a player
or the WPM for all 25 minute or longer games for a player
just an idea
great website!
|
|
|
the site says a typical grandmaster makes 3.7 workers per minute but wpm breakdown by league show 2-3 workers per minute for GMs. How should I understand this?
|
This is just amazing, I am using it already to improve my macro and keep track of it, thank you very much.
|
polybios, the article "Do you macro like a pro?" measures actual workers per minute, and says
In the first 28 minutes of the game, Grandmaster players produce an average of 3.7 workers per minute
The ggtracker site measures "worker waves" per minute, using a 3-second antispam filter, and uses the whole game, not just the first 28 minutes.
TheV, glad you like it!
|
This seems like a pretty bad measure for zergs since the consistency isn't there, even if it lacked the 3 second rule. "Time to get to X drones" seems like a better measurement, but even then it's hard to use since it can be so dynamic depending on opponent, there are many situations where it's downright stupid to drone too fast.
|
This is a useful/interesting site for a lot of people.
For me it's showed the same thing that the SQ statistic showed, I have macro which is well above the average for my league but I am lacking in other areas which holds me back. I've been in Plat for about a year now (in Diamond briefly) while my SQ was in the masters/GM range and using this my worker production graph is above the average Master line a lot of the time, and well above the Diamond line.
This confirms to me the same thing that I have felt for a long time, the game is far more complicated at a lower level than people give it credit for (especially on TL). The old maxim "if you build workers and build stuff you get masters ez" is simply not true.
I hope that this site can develop more and add more useful features. I'd like more choice on the graphs that are displayed, like the option to tweak which leagues it compares my graphs two, and also multiple players, e.g. choosing opponents/pros to compare our scores to.
|
On March 13 2012 00:00 RudePlague wrote:
This is a useful/interesting site for a lot of people.
For me it's showed the same thing that the SQ statistic showed, I have macro which is well above the average for my league but I am lacking in other areas which holds me back. I've been in Plat for about a year now (in Diamond briefly) while my SQ was in the masters/GM range and using this my worker production graph is above the average Master line a lot of the time, and well above the Diamond line.
This confirms to me the same thing that I have felt for a long time, the game is far more complicated at a lower level than people give it credit for (especially on TL). The old maxim "if you build workers and build stuff you get masters ez" is simply not true.
I hope that this site can develop more and add more useful features. I'd like more choice on the graphs that are displayed, like the option to tweak which leagues it compares my graphs two, and also multiple players, e.g. choosing opponents/pros to compare our scores to.
I agree, a tool to compare two players would be awesome, I am comparing myself vs Idra and clearly I have to make way more drones, his average on the reps is 2.3 and mine is 1.7 lol
|
On March 13 2012 00:00 RudePlague wrote:
This is a useful/interesting site for a lot of people.
For me it's showed the same thing that the SQ statistic showed, I have macro which is well above the average for my league but I am lacking in other areas which holds me back. I've been in Plat for about a year now (in Diamond briefly) while my SQ was in the masters/GM range and using this my worker production graph is above the average Master line a lot of the time, and well above the Diamond line.
This confirms to me the same thing that I have felt for a long time, the game is far more complicated at a lower level than people give it credit for (especially on TL). The old maxim "if you build workers and build stuff you get masters ez" is simply not true.
I hope that this site can develop more and add more useful features. I'd like more choice on the graphs that are displayed, like the option to tweak which leagues it compares my graphs two, and also multiple players, e.g. choosing opponents/pros to compare our scores to.
If you're truly on a masters level in both SQ and this worker production statistic, you should definitely write a post with replays asking for direction. Being way above your league in these statistics means that you always have way more stuff that your opponents, so there must be something fundamentally wrong in your play, which people can help you notice and correct.
|
RudePlague, great ideas, thank you! This project is definitely just beginning...
|
i have tried it, and it's actually very nice
two remarks to improve it maybe:
- is it complicated to give a tool to upload massively? (like I have 4000 replays, but couldn't upload them). I guess that would be huge, as it would give much more accurate statistics for everyone. But maybe it's hard to do
- more easily, maybe, when you break down the figures by league, I think you should mention not frequencies (like 7 % of masters have wpm at 1.7) but added frequencies (like 47% of masters have wpm at 1.7 or below). That would make more sense i feel, in the sense that you would be able to compare yourself to your league (like "more than 2/3rd of the players in my league have higher wpm than i do").
thanks anyway! I know I need to improve my macro now:D
EDIT: I also have a question, as a zerg player: How far from reality your proxy estimation of wpm is ? Cause I feel if the system actually only counts 1 drone per drone wave, it is quite problematic! especially after you made your first queen, you never really make drones 1 by 1.
|
Macpo, thanks for the compliments!
Good news! ggtracker already permits multiple replay upload. You can simply select multiple files when uploading, or you can drag-and-drop files onto the upload button.
Great idea about the by-league breakdown. We'll do that for the next version.
Regarding your last question, the current stat really measures drone waves per minute, and will be off by how many drones you make per drone wave. So if you typically make 2 drones per drone wave, then the ggtracker wpm stat will be 1/2 of your drone production. It's more accurate to think of it as measuring drone waves per minute. It's interesting to note that higher-league players do have more drone waves per minute. Will increasing your # of drone-waves improve your game? What do you all think?
Speaking of zerg, here are some stats that answer the question in another way, and address the question of those who wonder if worker-waves per minute are really related to win-loss. Looking at all the games in ggtracker, we can measure the worker-waves-per-minute (wwpm) for the winners separately from the losers. We exclude games < 5 minutes here.
- protoss: losers 1.84, winners 1.87
- terran: losers 1.74, winners 1.84
- zerg: losers 1.51, winners 1.50
I believe in the power of data! So until I or someone can dig up data that shows otherwise, I'll side with those who say that wwpm is not a relevant stat for zergs to focus on.
On the other hand, the wwpm difference for terran is pretty large. Digging into individual player stats, there's a ggtracker player with 387 terran games > 5 minutes, and their wwpm is 1.30 for their losses and 1.36 for their wins, with an average game duration of 22:10 and 21:56, respectively. Would individual player win/loss stats like that be relevant to helping us decide whether wwpm has an influence on winning and losing? Proving causation is hard to do without double-blind experiments, but we can gather lots of suggestive correlations...
|
I wonder, is there a connection between the two-hum distribution of SQ for bronze and the less pronounced two-hump distribution of wpm for bronze?
Also, I like the graphs, but I would be interested in seeing the cumulative distributions for each league. (Okay, so I just want to say my WPM is better than 95% of bronze league)
Edit; ninja'd by macpo.... I wonder how I found a page that old anyway.
|
89 matches played, 45 matches won (50.6%)
workers per minute trend 2.6 average, 2.9 recently
actions per minute trend 84 average, 85 recently
|
I love statistics as well, but one little concern, isn't your WPM higher lets say the first 12-15 min of the game? Since you try to max your worker count in the early stages. I would think that's where the big difference is between the two demographics.
Looking at the whole game length probably skews the results a bit, but would work in the favor of players not in master/gm. But then again, a master/gm is probably better at replacing lost workers during the course of the game which would then skew the graphs in their favor...
If this is the case, it would be nice to maybe look at the games in two parts, first 15 min, and then the rest of the game and maybe find some correlation to adjust the graphs.
just my 2 cents! Nice work!
|
|
|
|
|
|
EPT![[image loading]](http://i.imgur.com/kvzSX.png)
![[image loading]](http://i.imgur.com/NZGue.png)
