Advance multiplayer statistics

[Krom]

@Tom: "winners with big armies and few units lost" - army count can be a function of time (Pro usually have 100+ army at PT end) and loss ratio between players can kick in if armies were comparable (Pro manages army better than Amateur).

You can suggest your set of hints that identify Pro player

[T*AnTi-V!RuZz]

@Tom: "winners with big armies and few units lost" - army count can be a function of time (Pro usually have 100+ army at PT end) and loss ratio between players can kick in if armies were comparable (Pro manages army better than Amateur).

You can suggest your set of hints that identify Pro player

I think a pro player can only be identified by someone's playing style, something which is nearly impossible to measure.

[dicsoupcan]

If we collect the said data we can detect Pro players

how can you detect them if you do not collect nicknames?

Anyway i thought a bit about it and maybe it is funny to register the average time per game, maybe split between fighting and building and maybe even (if you want detailed stufF) betwene the speeds.

[Bence791]

Pro usually have 100+ army at PT end

Not 100+, but 70-80+ (assuming you mean 60 minutes, and valuable army, not militia rush).

@Dicsoupcan:
Collecting nicknames wouldn't be correct enough either, you can play any name and you'll stay Dicsoupcan (which the server couldn't tell)...

Also, high kill-loss ratio doesn't mean someone is a pro player, just imagine a situation in which 40 knights face 200 militia in a narrow space. The knights will win without much losses, and it gives a good ratio, right?
So personally I have no idea how it could be restricted to the better players, to be honest.

About the market: In the games I played with worse players (recently) only me (and other TS players) used markets, so I think it won't help with that question either.

[Ben]

If we collect the said data we can detect Pro players

how can you detect them if you do not collect nicknames?

Earlier in the topic, Krom said that people with large PT armies(/good bases?) will be considered pro.

[dicsoupcan]

If we collect the said data we can detect Pro players

how can you detect them if you do not collect nicknames?

Earlier in the topic, Krom said that people with large PT armies(/good bases?) will be considered pro.

But then you can only say hey someone got a large pt army, but you will not see whoever it was?

[T*AnTi-V!RuZz]

But then you can only say hey someone got a large pt army, but you will not see whoever it was?

That's not important for statistics.

[pawel95]

Earlier in the topic, Krom said that people with large PT armies(/good bases?) will be considered pro.

That´s also not a valid point to make statistics who is pro or not. When someone has a lategame base and 10 min after pt has a huge army, but his enemy has a big pt army,but after the first load nothing(because of market) it isn´t a sign to be a pro or not.

[Krom]

@Tom: if something can't be measured, then how do we know someone plays good and someone not?

@Dicso: if we collect names (which we dont want to do because we dont have accounts yet), who will judge if the name is a Pro or not, by which criteria, can it be formalized perhaps?

@Bence: We could assign scores to players army (militia = 1, knight = 6). Markets was just an idea, we will be able to look into it and see if there's anything suspicious (e.g. trunks > coal used 70% of the time would strongly hint to check that trade)

@Ben: I said that as an example, please don't consider it an utterly final decision xD


Please follow the topic: we are discussing WHICH stats might be interesting to collect along with some explanation WHAT for. Detecting if player was a Pro or not is a whole different topic AFTER we have those stats.

[Soldado con espada]

Let me refloat this old post...

The first time I saw @Esthlos trying to calculate how good/cost effective units are (see for example this for a method to calculate their cost), I thought... this could be done with Multiplayer statistics!

I see in previous posts that people were worried about "random players" destroying the statistics, and my only solution to this would be to record the nicknames. And I really am not sure why this would violate anyone's privacy...
Or, if you prefer, the nicknames can be encoded in a way that only the server admins know.

With this, you can start assigning weights to players and their performance.

And now, here comes the actual idea: count the number of units and which types, and the number of losses and unit types per player per game and store it. Then we could start comparing unit type "strength" against other units in combination with other unit types (yeah, a few dimensions to take into account xD).

If Player 1 creates 5 militias and 10 bowmen, Player 2 creates 3 axemen and 7 bowmen in total and Player 1 wins the game, you assign (for example) a weight of 1 to the first combination of units against the second. Then (somehow, I would need to think about it and certainly refine it with time) assign that weight to individual unit types* fighting with certain other unit types against other groups of units.
This would not give an accurate description of small battle outcomes, but an "overall" view.

Of course, skill and strategy of players is a very important factor... but to take this into account we would normalise the weight given to each game result using the player weight/expertise, calculated in... whatever way... taking into account win/lose, units produced/lost, etc, over time (this would require input from you, KaM experts, not me ). So for the first idea to work, it would be essential to record player ids, even if encripted



*so, for a very rough example that would require a lot of games to be played to refine it because it ignores the combinations of units... and maybe even after a lot of games it would never make much sense:
Bowman against axeman: weight += (1/P1weight)*(3/10)
Militia against axeman: weight += (1/P1weight)*(3/5)
Militia against bowman: weight += (1/P1weight)*(7/5)
Bowman against militia: weight -= (1/P2weight)*(5/7)

The weights would be updated taking into account every game outcome, and player weight could be calculated as a moving average or something like that...
In the end a table could be produced with each unit weight... as an average of all its weights, for example...

Very complicated, maybe not superuseful... but I would be happy to give it a try if I am given the data

[Soldado con espada]

There is an interesting side to this problem - if the game was collecting anonymised stats we could have had a huge base of games (like thousands) and with the right metrics we could see how many axes equal to how many pikes vs bows.

Not anonymous, keep the nickname to be able to apply weights in the future. If you are worried about anyone complaining, just include a warning/agreement where you tell the player that his/her nick will be stored.