Statistical analysis of in-game strategy
Andinagama
Join Date: 2012-12-23 Member: 176019Members
Join Date: 2012-12-23 Member: 176019Members
In spectator mode, you get to see exactly how a game progresses on both sides. The mode annouces to you, via the use of a graphical interface when events such as
'marines have researched armour 1' and
'aliens have researched regeneration'
'aliens have claimed Elevator Transfer for their 2nd hive'
'marines have claimed Server Room for their 2nd command station'
have occurred.
From a statistical standpoint, it would be interesting to know when which teams performed which strategies, and how well they worked (based on the end of the round results, and possibly also kills to some extent), to take the sum of all strategic 'events' logged by the game and perform statistical analyses on them, to answer questions such as
If marines spawn in Terminal at 0:00, which location is statistically more successful to expand to - Cafeteria, Locker Rooms, Departures?
What is the difference between Alien win percentages when they research a) carapace or b) celerity first (And then tell it to ignore all games where the first ability came later than 12 minutes in)
So my question is, is there some way to produce 'log files' from games, using a written program, in an output such as the following (for example)
Player names should be kept anonymous.
UWE Official EU #x Server
20:30 30/3/2012
Map: ns2_docking
Game Begins----
0:00 Aliens Spawn in Generator
0:00 Marines Spawn in Cafeteria
00:40 1_Skulk <bites> 1_LMG dead
00:43 2_LMG <shoots> 1_Skulk dead
1:00 Marines Research Phase Tech
1:35 Aliens Upgrade Crag Hive
.....
....
....
47:32 ARC <obliterates> Departures Hive
47:32 Marines Win, Aliens Lose
Then, from many, many such output files, we could run algorithms on them to answer strategy questions.
To create such an output file, one would need to identify variables in the game that are worth analysing (obviously my example above was not meant to be exhaustive).
Is this possible/desirable?
'marines have researched armour 1' and
'aliens have researched regeneration'
'aliens have claimed Elevator Transfer for their 2nd hive'
'marines have claimed Server Room for their 2nd command station'
have occurred.
From a statistical standpoint, it would be interesting to know when which teams performed which strategies, and how well they worked (based on the end of the round results, and possibly also kills to some extent), to take the sum of all strategic 'events' logged by the game and perform statistical analyses on them, to answer questions such as
If marines spawn in Terminal at 0:00, which location is statistically more successful to expand to - Cafeteria, Locker Rooms, Departures?
What is the difference between Alien win percentages when they research a) carapace or b) celerity first (And then tell it to ignore all games where the first ability came later than 12 minutes in)
So my question is, is there some way to produce 'log files' from games, using a written program, in an output such as the following (for example)
Player names should be kept anonymous.
UWE Official EU #x Server
20:30 30/3/2012
Map: ns2_docking
Game Begins----
0:00 Aliens Spawn in Generator
0:00 Marines Spawn in Cafeteria
00:40 1_Skulk <bites> 1_LMG dead
00:43 2_LMG <shoots> 1_Skulk dead
1:00 Marines Research Phase Tech
1:35 Aliens Upgrade Crag Hive
.....
....
....
47:32 ARC <obliterates> Departures Hive
47:32 Marines Win, Aliens Lose
Then, from many, many such output files, we could run algorithms on them to answer strategy questions.
To create such an output file, one would need to identify variables in the game that are worth analysing (obviously my example above was not meant to be exhaustive).
Is this possible/desirable?
Comments
Example game graphs:
http://ns2stats.org/round/round/71147
As for what exactly comes out, no clue.
@Mclawls, skill differentials certainly do exist.. that is why a very large number of games have to be recorded in this fashion, so that skill differentials have the best chance of balancing each other out. it will be a pretty useless tool of analaysis for a single game, but if 20,000 games could be recorded, then it has a higher chance of being statistically relevant.
It should of course also be possible to filter games by certain conditions - so if the marine team got, say, 5x as many kills as the alien team, disregard that game for this analysis point because of obvious skill balance (for example)
Yeah, this kind of stuff would be awesome to dig into. UWE certainly has the raw data that could be used to answer these kinds of questions (AND MORE!). I'm not certain they have the statistical expertise to get valid answers to useful questions out of the raw data, nor the database expertise to build the kind of data warehouse that would make that kind of data mining feasible. But they are very smart people, so maybe they do.
I'm pretty sure they don't have the TIME to do it, though. Pity. It would TREMENDOUSLY enhance their abilities to make the game more balanced and fun.
On the other hand, they're pretty obsessive about building tools, so maybe they've got this stuff at their fingertips already.
Someday they'll pull the curtain back a little further and we'll get to see for ourselves. They've talked a little bit about the database they use to store this kind of statistics ("Sponitor", aka Spark Monitor) and have talked about wanting to provide a public API for interacting with it, but that's probably a long, long way down the road, if ever.