EPT31,500 Replays - A look at balance in Starcraft II - Page 11
| Forum Index > SC2 General |
|
Jedclark
United Kingdom903 Posts
| ||
|
ZisforZerg
United States224 Posts
| ||
ZeromuS
Canada13379 Posts
On February 04 2012 15:02 lee365 wrote: Proxy 2 gate represent. 62% winrate might make it one of the most effective cheeses in the game I think the fact its used to so often in PvP vs. a standard opening is what is leading to it being the highest win rate vs standard P opening being one of the worst. Spurious events are awesome aren't they?  On February 05 2012 01:25 ZisforZerg wrote: who knew TDA was so balanced O>O Everyone did actually The only thing thats sucks about TDA is the 4 gate v 4 gate scenarios | ||
|
david0925
212 Posts
| ||
|
monitor
United States2403 Posts
| ||
|
astroorion
United States1022 Posts
| ||
|
Acritter
Syria7637 Posts
I don't think these statistics mean TOO much, though. They're over five months, and there were some pretty significant balance changes and metagame shifts in the past five months. | ||
|
Sleeples
United States8 Posts
| ||
|
malaan
365 Posts
Nice work. Work like this is always cool to see. | ||
|
Markwerf
Netherlands3728 Posts
For example tal darim has roughyl 1500 games in total, how many of those were PvZ? The best way to represent data like this is simple confidence intervals, that way you can see directly how valuable the data are and more importantly, if any differ significantly from 50%. With these amounts of games i suspect that anything over 52 (or below 48) is practically significant at 5% though so there is definately something going on. One more thing I'd like to see stats of are mirror matches. How often wins the player with more skill (measured by ladder rating or ladder win percentage in the particular matchup for example)? I''ve never seen someone look into this but mirrors DO affect balance overall and shouldn't be ignored. If a mirror is very luck based, let's say PvP for sake of argument, then it is more likely a good player get's knocked out by a worse player. Consequently the race will perform WORSE in other matchups as the mirrors don't act as a selection for better players. Vice versa, if the mirror features almost no luck, for sake of argument let's take TvT, then mirror matches will more consistenly weed out the worse players. As a result that race will perform BETTER in other matchups and skew the statistics in that way. I suspect that for a long time already P has worse stats and T has better stats because of this effect, which seems to be confirmed by the fact that top players tend to have worse PvP percentages then TvT ones. Overall I'd rather see balancing being done by changing the game (unit stats etc.) then by changing the maps. Balanced maps are often still bad because they don't produce interesting games, for example Tal darim may look very balanced but i'm not a fan of the map at all. PvP is terrible on it, the cliff at the natural means colossus or siege tanks attacks on it are very popular and PvZ is completely focussed around muta play on it. 50% maps are not the holy grail, fun maps with close to 50% are.. | ||
|
Integra
Sweden5626 Posts
On February 05 2012 01:58 malaan wrote: guess it's just me with a %20 winrate on taldarim TvZ then... :/ Nice work. Work like this is always cool to see. Keep in mind this is GM/higher masters/ pro players, The maps might not have the same win % for each race in your league. | ||
|
Thombur
95 Posts
On February 05 2012 01:59 Markwerf wrote: The best way to represent data like this is simple confidence intervals, that way you can see directly how valuable the data are and more importantly, if any differ significantly from 50%. With these amounts of games i suspect that anything over 52 (or below 48) is practically significant at 5% though so there is definately something going on. How are they supposed to give confidence intervals? The numbers give are the exact numbers they have. I see no real reason to try and use statistical analysis to find how big % of all starcraft 2 games being played are actually being played in the playhem daily and then how likely they are to represent the total. | ||
|
EndOfLineTv
United States741 Posts
| ||
|
Psylence
Canada118 Posts
| ||
|
dnld12
United States324 Posts
On February 04 2012 15:03 K3Nyy wrote: Nice. Even though it's not really at the highest level, it's still interesting to see. And lol at 2gate being Protoss highest win rate build order, while the standard being the lowest. ><"!! Thanks for this! Yep thats really sad T_T | ||
|
Smackzilla
United States539 Posts
On February 05 2012 01:22 Thombur wrote: Hot damn that's almost unbelievably well balanced imo, not to say there are no balance issues but still, cool statistics to see. Note that 5 moths ago = from september Last real patch was 20 September 2011 The 8 November 2011 made only minor balance changes (lower EMP radius, lower protoss upgrade costs) EMP radius is a pretty big deal. | ||
|
Vansetsu
United States1452 Posts
| ||
|
rOse_PedaL
Korea (South)450 Posts
| ||
|
Markwerf
Netherlands3728 Posts
On February 05 2012 02:04 Thombur wrote: How are they supposed to give confidence intervals? The numbers give are the exact numbers they have. I see no real reason to try and use statistical analysis to find how big % of all starcraft 2 games being played are actually being played in the playhem daily and then how likely they are to represent the total. the small problem with these numbers is that you don't know exactly how many games they are based on, confidence intervals are just the (generally agreed upon) clearest way to show the data and some insight in their reliability at the same time. You can easily make confidence intervals of these numbers based on a t-statistic, you don't really need to know how big % of all sc2 games are played in the daily as you can safely assume it's a irrelevant small number (thus you don't need to correct). The minor problem I have now is that I don't know how many games for example the numbers for tal darim are based on. I only know the total of those matchups plus the mirrors is around 1500.. Confidence intervals are easy to calculate and only provide more info.. | ||
|
nOlifeTERRAN
United States153 Posts
| ||
| ||
