|
On May 22 2013 21:26 EtherealDeath wrote: o.O interesting numbers, just checked my last 6 ladder games, half of which went to at least 3-4 bases, had a minimum uptime of 85% and a max of 90%... is 90% really the max uptime? Granted I kinda don't spread creep at all and prioritize injects for ling heavy style lol.
Because of how they measure, it would depend on how much energy your queens have before you start injecting. In theory, if you build several queens per hatchery and/or let them generate energy before the first inject on their "assigned" hatchery, you can reach 100% when it comes to inject uptime.
|
On May 22 2013 20:47 Stol wrote:Show nested quote +On May 22 2013 20:39 FLuE wrote:On May 22 2013 19:57 Stol wrote:Well I'm still waiting for just about anyone to actually reply to either of my two comments . Especially this segment in my second post: On May 22 2013 18:08 Stol wrote: I was also unable to determine if you rebalance the graphs to account for the fact that 90% is max. As an example, 66% out of 100 is still 66 but if you account for the fact that 90 is max, you actually end up at roughly 73.3%. 60% out 90 is on the other hand roughly 66.7. So instead of having a difference of 6 percentage units we're now sitting at a difference of 6.6 percentage units. An increase of the 10% we left out of the equation when using 100% as max instead of 90%.
Ofc when using more than one queen per hatchery it would then be possible to reach over 100% but I dont really think anyone is doing that. If you rescale the numbers putting 90% as the maximum uptime instead of 100%, the difference between the leagues increases. Edit: I'd prefer a reply to the other stuff as well though . For comparison sake it doesn't matter about that 10% because proportionally it will yield the same result. The graph would basically look the same, and the percent difference although as a raw number is higher it is out of 90 and not 100 so the difference is really the same. You would simply be manipulating numbers. Hopefully that makes sense and you see what you are suggesting would not make a real difference in data analysis. And for what it is worth I'd love if this type of analysis could be done on the percent of the map covered by creep. I think that is the data that would really show the difference in leagues. Or simply number of active tumors. There is a difference in the sense that what you would be measuring is the uptime of the actual injects available instead of the injects present on the hatchery. The fact that it would also affect every other aspect of the study along with standard deviation and so forth is irrelevant. The difference is in fact not the same as people are making their assumptions based on the difference in percentage points which in this case is smaller than the difference present when balancing the values. A point you could be making is people waiting with their first inject gaining extra energy before the injects occur which does have an impact on the possible larvae generated, but that is an entirely different matter and until the data also take into consideration when the queens themselves were spawned, one can only speculate on how much it would interfere with the end result. Edit: To make my point more clear lets visit a somewhat extreme value. Say that instead of 90%, only a 10% uptime was possible due to queen energy generation. A master player sitting at 8% would not look that much more impressive than a silver player sitting at 7% when we're disregarding the fact that 10% is the possible maximum and instead use 100%. It would only be a difference of 1 percentage point. However, seeing as 10% is the actual maximum we instead get the following result: 8/10 = 80% and 7/10 = 70%. He is in fact hitting his injects at a much more consistent rate. So while the amount of larvae generated isnt substantially higher, the better player is still a lot better at hitting his injects properly.
That's why it would be nice to have a proper test statistical test for the difference as well as I think it would indeed be quite significant. Instead the author screwed that up and compared the difference to the wrong standard deviation and made a weird statement implying the differences are pretty small.
This thread is pointless though for solid criticism, any good comments get buried beneath the sea of crappy statements.
Besides there are already far more extensive works out there for modeling player skill not even mentioning this method of relying on statistics to improve is very questionable. I'm a fan of using statistics for classifying players in sports especially those where a solid individual ranking doesn't exist like baseball but using using it to improve players themselves is a lot more dodgy.
|
Very interesting findings, however I think that the ratio between the inject% and number of hatches across leagues would be more valuable, though I am interested to hear your thoughts otherwise.
|
On May 22 2013 19:57 Stol wrote:Well I'm still waiting for just about anyone to actually reply to either of my two comments . Especially this segment in my second post: Show nested quote +On May 22 2013 18:08 Stol wrote: I was also unable to determine if you rebalance the graphs to account for the fact that 90% is max. As an example, 66% out of 100 is still 66 but if you account for the fact that 90 is max, you actually end up at roughly 73.3%. 60% out 90 is on the other hand roughly 66.7. So instead of having a difference of 6 percentage units we're now sitting at a difference of 6.6 percentage units. An increase of the 10% we left out of the equation when using 100% as max instead of 90%.
Ofc when using more than one queen per hatchery it would then be possible to reach over 100% but I dont really think anyone is doing that. If you rescale the numbers putting 90% as the maximum uptime instead of 100%, the difference between the leagues increases. Edit: I'd prefer a reply to the other stuff as well though . That's also an interesting way to look at it. However, wouldn't it be even more important to account for the theoretical minimum values? Since only hatcheries with at least two injects are included in the dataset, 0% is effectively impossible no matter what game we're looking at... and the actual minimum depends, first and foremost, on game length:
The graph starts at a game length of 5 game minutes. I dare say that if I have done two injects before the five minute mark, my injects are at least decent. (From the top of my head, if I open Hatch first, I'm not even sure if I could have my queens out early enough for two injects by the 5 minute mark in a best-case scenario.) On the other hand, if I messed up my injects, it wouldn't count because I had no hatcheries with at least two injects. In other words, it's hardly surprising to see silver players with >70% inject efficiency in short games when ~70% may be the effective minimum to reach the prerequisite of a twice-injected hatchery in time.
|
On May 22 2013 03:59 FawxzTV wrote: This mostly just shows that higher level players expand more aswell as adding macro hatches. Resulting in more bases than queens -> lower numbers. Injects are still INCREDIBLY important in the first couple of minutes. That's true. On top of that on a higher level your opponent will harrass more etc. Which will make you miss more injects.
|
On May 22 2013 04:27 mnck wrote: The reason highlevel players dont inject all the time is because they know when larva is needed and when creep is better. Sure maybe you could get more drones or army if you injected more, but the game is much more complex than that. Queens fight harass, spread creep and sometimes a queen per hatch instantly isnt the best. As for example any 3 base zerg build gets injects on 3rd hatch kind of late because otherwise you are at a surplus of larva but the 3rd base is still highly relevant for other reasons (Drone saturation and creep spread).
It's a good point that later game units require less larva and it can become less important. Also high level players can relatively easy achieved max (19) larva per hatch just from injecting lategame. This makes the queen float a lot of energy or makes injecting redundant.
Larva inject is by no means something that should be "kept up" but what would be a more interesting statistic, is how low the energy on the queens are. Thats something I'd love to see. If a queen stays low on energy then i'd be amazed! Not on the inject uptime.
This. This is the answer. I've seen players like Scarlett purposely miss injects because they feel they need better creep spread against what their opponent is going for. Also having insane numbers of larvae is pretty useless unless you are going for the fast remax.
|
On May 22 2013 22:16 Poffel wrote:Show nested quote +On May 22 2013 19:57 Stol wrote:Well I'm still waiting for just about anyone to actually reply to either of my two comments . Especially this segment in my second post: On May 22 2013 18:08 Stol wrote: I was also unable to determine if you rebalance the graphs to account for the fact that 90% is max. As an example, 66% out of 100 is still 66 but if you account for the fact that 90 is max, you actually end up at roughly 73.3%. 60% out 90 is on the other hand roughly 66.7. So instead of having a difference of 6 percentage units we're now sitting at a difference of 6.6 percentage units. An increase of the 10% we left out of the equation when using 100% as max instead of 90%.
Ofc when using more than one queen per hatchery it would then be possible to reach over 100% but I dont really think anyone is doing that. If you rescale the numbers putting 90% as the maximum uptime instead of 100%, the difference between the leagues increases. Edit: I'd prefer a reply to the other stuff as well though . That's also an interesting way to look at it. However, wouldn't it be even more important to account for the theoretical minimum values? Since only hatcheries with at least two injects are included in the dataset, 0% is effectively impossible no matter what game we're looking at... and the actual minimum depends, first and foremost, on game length: The graph starts at a game length of 5 game minutes. I dare say that if I have done two injects before the five minute mark, my injects are at least decent. (From the top of my head, if I open Hatch first, I'm not even sure if I could have my queens out early enough for two injects by the 5 minute mark in a best-case scenario.) On the other hand, if I messed up my injects, it wouldn't count because I had no hatcheries with at least two injects. In other words, it's hardly surprising to see silver players with >70% inject efficiency in short games when ~70% may be the effective minimum to reach the prerequisite of a twice-injected hatchery in time.
Yes, thats true, I should have thought of it myself. Depending on when the injects start compared to when the game ends, the gained uptime of injects on the overall population could be heavily overestimated as the games in which the players fail to produce the minimum amount of injects required are not included.
Edit: Even moving on into longer matches with more expansions this could be a problem, and possibly a part of the answer as to why the deviation within the leagues is so large. I would like to use the same example as was mentioned in the report, but with a slight twist.
For example, let’s say you had two hatches in a game that you won. The first one got its first inject at 4:45, and the second one at 7:45. You won the game at 10:45. So your first hatch was active for six minutes, and the second one for three minutes. So that’s nine total minutes that hatches were alive.
And let’s say you did six injects on the first hatch and three on the second. That’s nine injects total, and each inject lasts for 40 seconds. That’s six total minutes of injects being active. So your inject % for that game is 6/9 = 66.6%.
Lets use the same numbers but instead of winning at 10:45, you win at 11:45. Same amount of injects at first, you will ofc have a lower uptime but nothing fishy is going on.
Your first hatchery was active for 7 minutes and your second was active for 4. A total of 11 minutes of active hatcheries. With 6 injects on the first and 3 on the second you get the standard 9 injects in total. 6 total minutes of injects being active which means that 6/11 equals to roughly 0.545 so lets say 54.5%. Lets now say someone worse is playing in a similar scenario. 11 minutes of active hatcheries but they only manage 5 injects on the first and 2 on the second. 7 injects in total meaning 280 seconds of inject uptime. Over 11 minutes thats about 42.4% inject uptime.
Here it gets interesting. Say that instead of doing 5 + 2 you only do 5 + 1. The last hatchery is then not taken into consideration as it was only injected once. Your first hatchery was still active for 7 mintues and you managed 5 injects, 200 second uptime on injects over 7 minutes results in an uptime of about 47.6%. So someone injecting less than you can end up with a higher inject uptime. Now you can ofc play around with the actual duration of the games and the amounts of injects you manage to hit but the fact remains. Depending on when you expand and how fast you hit the injects on expansions, you can gain a lower score than someone far worse then you. In effect, the time the game ends compared to when you expand plays a larger role on the result than your ability to inject properly.
|
Inject % = (total # of minutes all hatches spent with injected larva) / (total # of minutes all hatches were active)
So where do the number of hatches factor in here? Masters usually can manage more bases as the game progresses. I hate pretentious number crunching. >
|
Not a big fan of your statistical analysis - PLEASE differentiate between percent and percentage points! What is your standard deviation in for example?
|
On May 22 2013 22:42 TheFish7 wrote:Show nested quote +On May 22 2013 04:27 mnck wrote: The reason highlevel players dont inject all the time is because they know when larva is needed and when creep is better. Sure maybe you could get more drones or army if you injected more, but the game is much more complex than that. Queens fight harass, spread creep and sometimes a queen per hatch instantly isnt the best. As for example any 3 base zerg build gets injects on 3rd hatch kind of late because otherwise you are at a surplus of larva but the 3rd base is still highly relevant for other reasons (Drone saturation and creep spread).
It's a good point that later game units require less larva and it can become less important. Also high level players can relatively easy achieved max (19) larva per hatch just from injecting lategame. This makes the queen float a lot of energy or makes injecting redundant.
Larva inject is by no means something that should be "kept up" but what would be a more interesting statistic, is how low the energy on the queens are. Thats something I'd love to see. If a queen stays low on energy then i'd be amazed! Not on the inject uptime.
This. This is the answer. I've seen players like Scarlett purposely miss injects because they feel they need better creep spread against what their opponent is going for. Also having insane numbers of larvae is pretty useless unless you are going for the fast remax.
Ya, people are clinging too hard on their 'injects measure how good you are' rule of thumb and confusing the study as saying something it's not.
All it's saying is that there's no strong correlation between inject uptime (even when taking out things like Macro hatches) and player-performance. It even says there's a strong correlation between APM and Spending Quotient so starts to speculate for why the disconnect. I don't think they hit all the right reasons, but they hit a few good ones.
A lot of people are right when they say Queen Energy is what's actually important (since queens can do other things than inject) and there are builds where you just don't inject non-stop *even* at the beginning of the game. IE - 3 hatch before pool, you do not have the minerals for all that larva and unless you game the metric by only injecting on the same hatch (possible) you will end up with a poor inject uptime on active hatches (that's a mouthful).
My take-away is that injects should be viewed instead of an essential benchmark, like hitting your overlords, and more of a powering mechanic, like adding extra rax or gateways.
Of course it's free for Zerg to do this, so in an ideal world they would keep up injects late game just in case (how many times has a ling / roach remax failed because the player didn't have the larva for it?). On the other hand, adding Macro hatchings and using more Larva efficient units (never seen an Ultra remax fail for larva...) *do* make injects less important late game, so it's still not the end-all-be-all of Zerg mechanics.
|
On May 22 2013 04:57 Kaitlin wrote:Show nested quote +On May 22 2013 04:49 Teoman wrote: I think what some people are saying is really what it is.
I don't think higher level players are just as bad at injecting as lower level players. It is just that there are a lot more factors that make constant injects impossible at higher levels (harrass, difference in base taking, emphasis on creep, idle larva). Maybe part is the fact that higher level players are active on the map, while lower levels are looking at their bases lol.
I think the difference then would be: Lower level players look at their base or at the field. high level players do both.
|
Based on the article linked, we can see that progamer zergs consistently hit their inject % at 10-15% higher than masters level zergs for the same game-lengths.
Hence, injects are meaningfully correlated with skill- just not as much as things like apm...
|
You made it into This Week in Starcraft 2! :D
|
This study confuses me. I hate to go against so much data that seems well analyzed. All the stats i've studied tell me to trust the data, but this is as counter intuitive as it gets. All of my anecdotal evidence points to the norm statement that injects are absolutely crucial.
With that being said, my current opinion of this data is that your percentage, no matter how small, must make a monstrous difference. 5.1% improvement between masters and silver, must be one of the pillars that allow players to advance leagues. I think i agree with a large number of people's opinions in here. I believe the most useful statistic would be the total number of larva per unit time of the players of each league. 5% more larva must be enough that zerg is able to maximize drone production, and put out just enough units to hold some of the timings that come, that would kill less injectfully skilled players. Is there anyway we could find out something along the lines of how much more injected larva are produced by master players than silver players per minute or time. I am then curious to see if its like 10 more larva at ~9 minutes or something. Because then I can see master players surviving timings that would kill lesser players. I just repeated myself. Whatever.
Anyway, neat study, as with all suprising findings I would analyze more data a different way, and see if you come up with the same result.
|
On May 22 2013 21:29 Stol wrote:Show nested quote +On May 22 2013 21:26 EtherealDeath wrote: o.O interesting numbers, just checked my last 6 ladder games, half of which went to at least 3-4 bases, had a minimum uptime of 85% and a max of 90%... is 90% really the max uptime? Granted I kinda don't spread creep at all and prioritize injects for ling heavy style lol. Because of how they measure, it would depend on how much energy your queens have before you start injecting. In theory, if you build several queens per hatchery and/or let them generate energy before the first inject on their "assigned" hatchery, you can reach 100% when it comes to inject uptime.
you can never get 100% perfect inject uptime because your first hatch always takes 50sec of not being injected while your queen is on the way + 60sec for the pool, at least.
|
On May 22 2013 22:22 Vandrad wrote:Show nested quote +On May 22 2013 03:59 FawxzTV wrote: This mostly just shows that higher level players expand more aswell as adding macro hatches. Resulting in more bases than queens -> lower numbers. Injects are still INCREDIBLY important in the first couple of minutes. That's true. On top of that on a higher level your opponent will harrass more etc. Which will make you miss more injects. These two things seem like obvious flaws in the testing method.
|
On May 22 2013 03:54 Embir wrote: Finally solid confirmation that Zergs macro is the easiest - we already knew they had it easy with only one production building and easiest tech switches in the game, now we know that they macro mechanic is also forgiving - and note that supposed unforgiveness of zerg mechanics was main argument for zerg's macro difficulty.
Haha, thanks for being "that guy", saying what everyone was thinking.
|
This study is useless as you guys didn't control for the number of hatcheries the players have. A masters player is far more likely to end the game with more hatches than a gold player. A better study would have been to examine pro level games and check how inject percentages affect win rates, preferably by taking a look at many multi-game series between two opponents.
|
On May 26 2013 10:46 kckkryptonite wrote:Show nested quote +On May 22 2013 03:54 Embir wrote: Finally solid confirmation that Zergs macro is the easiest - we already knew they had it easy with only one production building and easiest tech switches in the game, now we know that they macro mechanic is also forgiving - and note that supposed unforgiveness of zerg mechanics was main argument for zerg's macro difficulty.
Haha, thanks for being "that guy", saying what everyone was thinking. Just wait until they do a study that says you can not drop a mule till full energy and still be fine. :/ Actually nevermind don't really need a study for that. Zerg macro is more than injecting, not to mention it takes more to inject 4+hatcheries than 1-2 of lower leagues. This study has nothing to do with the difficulty of it either, just saying you don't have to have every hatch constantly injected to win. If you understood Zerg macro more you would probably already know that, it is about balancing larvae, supply and income.
|
very simple find but i love it.
resources like this give us the information we need to focus on what is really happening and what is really needed in play.
cheers.
|
|
|
|