TvZ/ZvP/PvT stats on maps over time - Page 3

Milkis

5003 Posts

March 29 2011 02:04 GMT

#41

On March 29 2011 06:27 Necosarius wrote:
I think it's pretty cool how the graphs evens out on almost all maps (ofc there are some exceptions. Polaris rhapsody lol.)

It's not really "evening" out. It's the flaw of it being over time and not over number of games -- there are times where the "balance" doesnt change because well, no games were played during that time.

Grobyc

Canada18410 Posts

March 29 2011 02:09 GMT

#42

Python for Protoss... ewww =/

puppykiller

United States3137 Posts

March 29 2011 02:12 GMT

#43

On March 29 2011 06:31 dRaW wrote:
I've always hated PvT on Python, I found it quite hard because the play-style is much different than newer macro maps (far third and very open, circuit breakers is a little more similar play-wise). I find it funny how the stats show it that terran favoured when so many terrans complain that it's protoss imba

Imba for toss at low level since aggresion is so nessecary for tvp on that map and fails with noob terran apm and awful push micro

Carefree

United States1571 Posts

March 29 2011 02:33 GMT

#44

On March 29 2011 11:04 Milkis wrote:

Show nested quote +

Why don't we just do it by game then?

Release

United States4397 Posts

March 29 2011 02:33 GMT

#45

On March 28 2011 17:33 Milkis wrote:

Show nested quote +

Also Polaris Rhapsody
[image loading]

screw that map

Funny how jaedong never thumbed down Polaris Rhapsody as the dual map.

salito

1647 Posts

March 29 2011 02:34 GMT

#46

Well it was Polaris or Odd-Eye, neither a favorable choice.

Milkis

5003 Posts

March 29 2011 02:55 GMT

#47

On March 29 2011 11:33 Carefree wrote:

Show nested quote +

Why don't we just do it by game then?

Cause I haven't coded it that way yet, silly!

But graphing by time also has its advantages in that you get to see exactly at what times the shifts occur. That itself makes it interesting to look at.

I will address some of the other stuff tomorrow, and I guess I could start a thread on some stat theory cause I think that's interesting hehe. But that will have to wait till tomorrow cause I have to translate like 15 pages tonight apparently.

Chimpalimp

United States1135 Posts

March 29 2011 03:58 GMT

#48

This should be used in the ISL for sure.

Crunchums

United States11144 Posts

March 29 2011 04:09 GMT

#49

On March 29 2011 05:41 xarthaz wrote:

Show nested quote +

If you reject the premise of player skill being equal for all three races(which is the reasoning for not including WL or ace matches or whatever) then that also implies that no statistics at all can reflect the map balance

What? Even if you reject that premise statistics can still reflect map balance - you just need to compare against a baseline that reflects the difference in skill between the three races.

The reason WL doesn't make sense is because I am trying to use usage statistics. In regular proleague non-ace matches every time a player is sent out they were specifically chosen for that map, and hence if a map favors a certain race you ought to see that race be sent out more frequently than you would expect if it were perfectly balanced. In winner's league it doesn't say much that crazy-hydra played on circuit breaker if the reason he did that was he happened to win 3 games before that. Maybe you could argue that you could still analyze the first game a player is sent out for but when Jaedong gets sent out in the 4th set that's not because of the map that's because god damnit jaedong oz is down 0-3 again. The same reasoning applies to ace matches in regular proleague - you would expect the correlation between the race that is sent out for an ace match and how good the map is for that race to be lower than non-ace matches.

edit: Did I just get trolled? I just spent a lot of time typing out some stuff that I think is pretty obvious.

hauton

Hong Kong743 Posts

March 29 2011 04:36 GMT

#50

Icarus/Aztec?

4vvhiplash7

South Africa392 Posts

March 29 2011 04:41 GMT

#51

On March 28 2011 17:26 Milkis wrote:
Fighting Spirit
[image loading]

Fighting Spirit was seriously one of the best maps in my opinion, they should bring it back...

konadora

Singapore66358 Posts

March 29 2011 15:12 GMT

#52

wow FS.. DAT BALANCE.

Divine-Sneaker

Denmark1225 Posts

March 29 2011 15:25 GMT

#53

Fucking polaris rhapsody. That was such a terrible map with a bunch of drawn out games or games that were only won due to the map imbalance.

Cool statistics.

aimaimaim

Philippines2167 Posts

March 29 2011 16:34 GMT

#54

Okay, I'm confused.

PvTwinration means Protoss wining against Terran or is it vice-versa?

So, the Python's Graph indicates a huge favor for Terrans against Protoss?

revy

United States1524 Posts

March 29 2011 17:45 GMT

#55

Milkis, could you possibly show us the first derivative with respect to time of these graphs (You'll likely need to smooth the plots or else they'll look like garbage) or a sliding window average?

That would help to show trends in the matchups at a given time. For instance destination was initially pretty imba for TvZ but sometime between december 2008 and march 2009 it started turning around and zergs began winning more often which eventually brought the MU down to ~55%.

If you have questions about how you might go about it you can PM me, I do these sorts of things for a living (though am out sick today!).

Milkis

5003 Posts

March 29 2011 18:07 GMT

#56

On March 30 2011 02:45 revy wrote:
Milkis, could you possibly show us the first derivative of these graphs (You'll likely need to smooth the plots or else they'll look like garbage) or a sliding window average?

That would help to show trends in the matchups at a given time. For instance destination was initially pretty imba for TvZ but sometime between december 2008 and march 2009 it started turning around and zergs began winning more often which eventually brought the MU down to ~55%.

If you have questions about how you might go about it you can PM me, I do these sorts of things for a living (though am out sick today!).

First derivative of a percent change would be rather misleading, so it's not a very good measure. You can see why cause as the number of games increase the average has less and less variance. I'll probably do a 10 game running average maybe in order to see noticeable trends.

I have no idea how to smooth plots though, so I'd be interested in learning how to do that since I have some very interesting plots that I will reveal soon

Also if you do this for a living I envy you ;___;

hmmm doesn't that mean sort of pull the statistics to extremes? if one map is really imba, then the stats will show it with a huge disparity in win rate, since there's a huge disparity + small sample size. however if a map is balanced, then there will be ~50% win rate + large sample size. so there's an exponential increase in "balance" as w/l stats approach 50%...? so assuming what i just said is true, FS is SUPER DUPER balanced? ^^

It biases statistics, yes.

I'm not talking about winrates, of course -- cause again I don't think winrate is an accurate measure of balance ("whatever that means"). There needs to be a good definition of balance before we even start talking about anything.

What about an aggregate racial balance chart, so we can see overall map balance over time (or was that already done before)?

Will be done, with a lot more, in a future post

Winner's league games should be thrown out when considering usage rates for obvious reasons. I would also throw out all ace match games as well; while individual games might support a certain argument (eg KT not sending Flash for an ace match on a map support that being bad for terran) I don't think KT sending Flash or Oz sending Jaedong on a certain map says much about the map's balance.

You can also account racial differences inherent in the scene by not just comparing the data to a naive baseline of 50%. The data will actually be more useful not that matches are Bo7 too, which is nice.

There are still flaws (eg if a team gets 4-0'd we lose 2 games) but if you are doing a statistics based analysis of map balance I think it is the best approach.

Yeah, I think it's also fair to assume it will even out in the long run. I would however have to filter out ace matches.

bITt.mAN

Switzerland3693 Posts

March 29 2011 18:56 GMT

#57

Gee this is really cool to look at, and somehow, against my heart I am now respecting Fighting Spirit a bit more as a map because of it's balance. I mean, all you need to do is macro like a cow and it's all good ; ] JK JK don't kill me (this is from a ~~guy~~

who loves LT, 815, Plasma and Monty Hall soo...)

About TvP @ Python, we need to remember that these are PROgamer statistics, not D P v D+ T stats (I still hate TvP on Python, so Toss imba yo!) because as has been said low-level Terran versus Low level Protoss is already unfair enough, and Python just makes it harder.