EPTDo you macro like a pro? - Page 46
| Forum Index > SC2 General |
|
The_Austin
United States6 Posts
| ||
|
StarStrider
United States689 Posts
| ||
|
AySz88
United States83 Posts
| ||
|
icarly
United States400 Posts
Too bad I use 2 control groups, have 70ish effective apm, am completely uncomfortable with ghost usage, and dont use camera saves :'( | ||
|
Flew
United Kingdom133 Posts
I'm silver terran. I measured my SQ over 50 games and I found that my average was 46, which puts me in the typical silver range. I plotted my results on a graph and added a best fit line: ![[image loading]](http://i.imgur.com/frO6f.png) You can see that there's a lot of variation (from -2 to 82), and I should probably include some error calculation, but I'm not sure how. However, the pleasing thing about this is that although I obviously need to improve, the general direction is up! I wanted to do a little investigation into how my SQ affected my winrate. For 50 games, I recorded my SQ and my opponent's SQ, and whether I won or lost. I discovered the following: 1. My overall winrate for the 50 games was 50%. 2. I had a higher SQ than my opponent in 50% of the games. 3. My average SQ was 46, my opponents' average SQ was 45. 4. For games where I had the higher SQ, my winrate was 68% 5. For games where my opponent had the higher SQ, my winrate was 32%. *Games where either myself or my opponent had less than 600 income were excluded. I plotted wins and losses on a graph like so, comparing my SQ to my opponent's SQ: ![[image loading]](http://i.imgur.com/7jLBo.png) If you imagine a diagonal line splitting the graph equally from bottom left to top right, then results in the top left are where I had the higher SQ, and bottom right is where the opponent had a higher SQ. Ideally we would like to see more blue results (wins) in the top left diagonal, i.e. person with highest SQ wins. You can see that there are some results where my SQ was far higher than my opponent's yet I still lost. I hate losses like this, and it usually results in me having an IdrA-like tantrum and thinking, "but I should have won, I was the better player!". As others have pointed out though, macro is just one part of the game, and if I'm focussing hard on my macro, I'm probably neglecting something else. Overall though, you can see that for my last 50 ladder games, a higher SQ seems to equate to a higher winrate, and the fact that whatthefat's results show that SQ goes up by league proves that a high SQ is definitely something I should be aiming for. I would have liked to do a similar analysis of SQ vs winrate for IdrA, but as I'm on the EU servers I had to choose someone else. I went for Thorzain, as he's a terran that I admire, and his reputation as the "Spoon killer" made me hope that he would have lots of lovely long games in which to measure his SQ. The graph for Thorzain's last 50 ladder games looks like this (no custom games were included): ![[image loading]](http://i.imgur.com/GPhFD.png) Again, games where one player had less than 600 income were excluded. Thorzain's winrate was 82% His average SQ was 94!! His opponents' average SQ was 80. Thorzain had a higher SQ than his opponent in 47 of the 50 matches. When Thorzain had the higher SQ, his winrate was 81%, where his opponent had the higher SQ, his winrate was 100%. As Thorzain had so few losses and so few examples of games where he had the lower SQ, I feel that this isn't really a good sample, but nonetheless, it was interesting to discover what a beast Thorzain is on ladder. I will continue recording my SQ, and hopefully it will continue to rise. I cannot emphasize enough what a useful tool this is. I know I need to work on my macro, and it's great to have a tool which helps to measure how I'm doing. Btw, the game where I had a -2 SQ was probably caused by me sitting outside my opponents base for 30 minutes, with 7k minerals in the bank. I suck at TvT. | ||
|
figq
12519 Posts
Wondering what Thorzain does, when he has contained the opponent like you in that situation. I guess he keeps poking and advancing; +reinforcing. He's actually incredibly active with his army at all times. Note also the main difference between an average ladder player and a pro is the consistency. | ||
|
tGFuRy
United States537 Posts
| ||
|
whatthefat
United States918 Posts
On September 27 2011 03:07 Flew wrote: I feel like I'm a little late to the party here, but I wanted to add my thanks for a great post. I'm silver terran. I measured my SQ over 50 games and I found that my average was 46, which puts me in the typical silver range. I plotted my results on a graph and added a best fit line: You can see that there's a lot of variation (from -2 to 82), and I should probably include some error calculation, but I'm not sure how. However, the pleasing thing about this is that although I obviously need to improve, the general direction is up! I wanted to do a little investigation into how my SQ affected my winrate. For 50 games, I recorded my SQ and my opponent's SQ, and whether I won or lost. I discovered the following: 1. My overall winrate for the 50 games was 50%. 2. I had a higher SQ than my opponent in 50% of the games. 3. My average SQ was 46, my opponents' average SQ was 45. 4. For games where I had the higher SQ, my winrate was 68% 5. For games where my opponent had the higher SQ, my winrate was 32%. *Games where either myself or my opponent had less than 600 income were excluded. I plotted wins and losses on a graph like so, comparing my SQ to my opponent's SQ: If you imagine a diagonal line splitting the graph equally from bottom left to top right, then results in the top left are where I had the higher SQ, and bottom right is where the opponent had a higher SQ. Ideally we would like to see more blue results (wins) in the top left diagonal, i.e. person with highest SQ wins. You can see that there are some results where my SQ was far higher than my opponent's yet I still lost. I hate losses like this, and it usually results in me having an IdrA-like tantrum and thinking, "but I should have won, I was the better player!". As others have pointed out though, macro is just one part of the game, and if I'm focussing hard on my macro, I'm probably neglecting something else. Overall though, you can see that for my last 50 ladder games, a higher SQ seems to equate to a higher winrate, and the fact that whatthefat's results show that SQ goes up by league proves that a high SQ is definitely something I should be aiming for. I would have liked to do a similar analysis of SQ vs winrate for IdrA, but as I'm on the EU servers I had to choose someone else. I went for Thorzain, as he's a terran that I admire, and his reputation as the "Spoon killer" made me hope that he would have lots of lovely long games in which to measure his SQ. The graph for Thorzain's last 50 ladder games looks like this (no custom games were included): Again, games where one player had less than 600 income were excluded. Thorzain's winrate was 82% His average SQ was 94!! His opponents' average SQ was 80. Thorzain had a higher SQ than his opponent in 47 of the 50 matches. When Thorzain had the higher SQ, his winrate was 81%, where his opponent had the higher SQ, his winrate was 100%. As Thorzain had so few losses and so few examples of games where he had the lower SQ, I feel that this isn't really a good sample, but nonetheless, it was interesting to discover what a beast Thorzain is on ladder. I will continue recording my SQ, and hopefully it will continue to rise. I cannot emphasize enough what a useful tool this is. I know I need to work on my macro, and it's great to have a tool which helps to measure how I'm doing. Btw, the game where I had a -2 SQ was probably caused by me sitting outside my opponents base for 30 minutes, with 7k minerals in the bank. I suck at TvT. Thanks Flew, very nice! I'm actually looking to expand this analysis to include pros from other servers in the near future. If you're interested in collecting some data from the European server, PM me. I welcome input from others on non-American servers too. | ||
|
Flew
United Kingdom133 Posts
On September 27 2011 03:47 figq wrote: ^ Thanks for that post. Curious that your lowest SQ (-2) comes from a game you were winning. So it's also interesting to analyze the second plotting only in terms of your own SQ and winning. I agree, and it's also interesting to see the game where my SQ was over 70, my opponent's was just over 40, yet I still lost! I probably shot to max pretty quick, then walked my army into a tankline / mass of banelings or something. I will definitely be downloading some of Thorzain's replays and trying to pick up some tips! I always seem to get a big pile of unspent minerals while I'm trying to establish my first expansion, so I'm going to try see what the pros do around that time. I'm obviously not putting down production buildings fast enough I guess. On September 27 2011 04:19 whatthefat wrote: Thanks Flew, very nice! I'm actually looking to expand this analysis to include pros from other servers in the near future. If you're interested in collecting some data from the European server, PM me. I welcome input from others on non-American servers too. Thanks for the feedback! I'd be happy to help with data collection. I must say, after recording results from just 100 games from battlenet I was starting to get a headache, so I have a huge amount of respect for the patience it must have required to go through 2000 games! I'm interested to see how far this anaysis can go, and I think it would be great to include lots of things, it's just a shame that we're limited in the data we can get from replays. Do you think it's worthwhile recording opponent's SQ as well? Do you have any other plans for how SQ could be used in other studies? | ||
|
The_Austin
United States6 Posts
I'm posting this because the excel file I created is a large improvement from the file given and includes a lot more information if anyone wanted an easy way of figuring there own stats out. The file I included has my information in it but you only need to input Average Unspent and Collection rate for each game and the rest of the info will fill out. Here's an image. ![[image loading]](http://img26.imageshack.us/img26/1893/sc2quotient.jpg) And here's the file if you're interested! Enjoy! http://www.mediafire.com/?c3gchw5a389t9sa | ||
|
Velvet_Llama
United States25 Posts
I had a few random thoughts I'd like to share with you. With regards to the relationship between number of workers and game duration, I'm wondering whether there is a curvilinear relationship. It looks like might actually be a basic linear relationship but with non constant error variance. Often times heteroscedasticity can screw with curve fitting. It might be the case that the length between worker production and game length is linear, but early limiting factors, such as low supply and the need to constantly produce probes might limit variability in the early game, while later game probe production varies more with different strategies and situations. The S-curve is still reasonable and might be the proper fit if there is a period of maximum probe production proceeded and followed by lower production. With regards to grandmaster players spending in an exponential fashion, might it be that, as a typical game progresses, income increases non linearly as you take more bases and produce more probes, while the "opportunities for spending" and the rate at which one can actually spend remains relatively constant or only increases linearly? Particularly with Grand master level players we might expect that they're basically spending at peak efficiency from the very beginning of the game. Perhaps the amount of "stuff" you can buy just doesn't keep pace with the available income? I think the amount of time spent managing resources probably also slows down production a bit as the game progresses, regardless of level. Just a thought. With regards to the distribution of SQ- those are very normal (Gaussian) distributions. They look very "well behaved"- a sure sign of a good metric  On an unrelated note, I'd actually be surprised if skill of the overall player base followed a Gaussian distribution. Typically something like that would be expected to follow an "extreme" distribution, such as a Gamma or Weibull (note that Weibull distributions show up all over the place in Sabermetrics for the same reason). I know it would take another annoyingly long period of data collection, but the next logical step would be to see how well your SQ predicts the league of a player. An ordered probit model would probably work well. But again, obviously that would require you to grab a new sample which was a pain in the ass Finally, for future reference, when doing curve fitting with empirical data, I always like to start with non parametric fits, like a spline based fit or a lowess smoother. Often times you can get a good sense of how the data are behaving and whether they obviously appear to fit a given type of curve. It just feels a little less arbitrary than deciding "well that kind of looks like a lognormal distribution." If you use the analysis package "R," (and if you aren't, you should :p ) there are some really nice packages for non parametric exploratory analysis. Anyways, awesome work man. | ||
|
Frightmare
46 Posts
Many peope have already noted in lower leaugues people don't build enough workers which inflates their sq. But what hasn't been mentioned is that at higher levels of play you are likely to be put under a lot more pressure which is gauranteed to decrease your sq. This explains the people who say they're in platinum but have 70+ but don't know why. Its because when your mmr gets high enough to play people who know how to pressure you, your sq probably drops considerably. For example I find I win against terrans when my mutalisk harass is doing tons of pressure supply blocking etc and causing him to float money. However the math here is insightful and well done. I would be interested to see if someone crunched the numbers of income vs time across leagues as well. | ||
|
No_eL
Chile1438 Posts
u are engineer or mathematician?? | ||
|
Damnight
Germany222 Posts
Yeah.... no. He also lost some, thats why that number is THAT high. Saturating is between 80 and 100. | ||
|
whatthefat
United States918 Posts
On September 28 2011 05:17 Damnight wrote: "saturating at 132 workers" Yeah.... no. He also lost some, thats why that number is THAT high. Saturating is between 80 and 100. I'm not saying they necessarily ever have 132 workers on the board. When I say "saturating", I'm referring to the saturation value of the fitted sigmoid curve. In other words, for long games, the average Grandmaster level Zerg player approaches a total of 132 workers constructed per game. | ||
|
Izanagii
Spain14 Posts
Great work with this. You're a gentleman and a scholar. | ||
|
habermas
United Kingdom304 Posts
| ||
|
Azera
3800 Posts
| ||
|
Kyamo
Canada129 Posts
This data is over 10 games. My SQ (protoss) = lowest 43.4, highest 67.8, average 56.4 Partner's SQ (zerg) = lowest 42.6, highest 79.5, average 57.8 |Partner's SQ - My SQ| (absolute value of the difference) = lowest 0.4, highest 11.7, average 6.29 Although we had individual ranges of 24.4 and 36.9, we never had an absolute value of the difference greater than 12, and usually more like 6. Plotting my scores against his scores, a linear fit gives an R^2 of 0.6692. I know this isn't a very big sample, but I think it definitely suggests the variability is more due to circumstances of individual games as opposed to just natural variation in performance from game to game. For all of these games we used the same (defensive) opening, then reacted to what was going on. No cheese included. No resource trading occurred in these games. | ||
|
Sunrunner
United States80 Posts
| ||
| ||
