Problems registering at AoCZone?
You can try resending activation email. If that doesnt work you can send an email here. If you forgot your password click here.

[KotD] Clash of the Civilizations (Stats)

Moderator:  swissboyy

Advertisement from Google 
 

Re: [KotD] Clash of the Civilizations (Stats)

Postby  philipp94831 » Tue Nov 28, 2017 8:46 pm

  28 Nov 2017, 19:08 GMT » Rey_Fer wrote:
Could you explain more extently how does your agorithm work and what data you took into consideration?


The idea is to construct a graph where each civ is a node. The nodes are connected via directed edges. There is one edge between two civs A and B if A lost to B. The number of losses determines the weight of an edge. We normalize the weights such that the sum of weights of all outgoing edges is 1 for each node.
After we initialized the graph, we want to model the probability to reach a node if we randomly walk the graph. We start at a random node and then either visit another random node or visit one of its neighbors. The probability to visit any node randomly is uniformly distributed. The probability to visit one of the neighbors depends on the weight of the edge between them. From that, we can derive the probability to randomly visit a node which is equivalent to the probability that a random civ we lost to is X.

More information: https://en.wikipedia.org/wiki/PageRank
 philipp94831
 
Posts: 20
Joined: Mar 23, 2017
 
 

Re: [KotD] Clash of the Civilizations (Stats)

Postby  Rey_Fer » Tue Nov 28, 2017 9:07 pm

  28 Nov 2017, 19:46 GMT » philipp94831 wrote:
  28 Nov 2017, 19:08 GMT » Rey_Fer wrote:
Could you explain more extently how does your agorithm work and what data you took into consideration?


The idea is to construct a graph where each civ is a node. The nodes are connected via directed edges. There is one edge between two civs A and B if A lost to B. The number of losses determines the weight of an edge. We normalize the weights such that the sum of weights of all outgoing edges is 1 for each node.
After we initialized the graph, we want to model the probability to reach a node if we randomly walk the graph. We start at a random node and then either visit another random node or visit one of its neighbors. The probability to visit any node randomly is uniformly distributed. The probability to visit one of the neighbors depends on the weight of the edge between them. From that, we can derive the probability to randomly visit a node which is equivalent to the probability that a random civ we lost to is X.

More information: https://en.wikipedia.org/wiki/PageRank


Do you have enough data in this tournament to make that graphe?
 Rey_Fer
 
Posts: 18
Joined: Feb 02, 2016
 

Re: [KotD] Clash of the Civilizations (Stats)

Postby  Michaerbse » Tue Nov 28, 2017 9:10 pm

  28 Nov 2017, 11:48 GMT » green eggs wrote:
(...)

I agree with a couple of previous posters that the Elo ranking system isn't super meaningful at the moment as it could be skewed by skill differences between the players in the tournament's early rounds.

At the moment this is just an issue of sample size. It would be really cool if this ranking system could become a "living document" that gets continually updated across multiple 1v1 tournaments. With time these stats can become really valuable (and can be a more objective indication of where the game might need rebalancing).

Problem being that civs' strengths are very different depending on the map type and not all tournaments are played solely on Arabia. E.g. having one single rating for the Italians, which considers both land and water maps, would not be meaningful.
You can create one rating per map category, e.g. one for "standard" land maps, one for arena (maybe + fortress, hideout, ...), one for nomad, one for water maps... But this will make it very difficult to reach a sufficient sample size again.
 Michaerbse
 
Posts: 15
Joined: Oct 14, 2017
 

Re: [KotD] Clash of the Civilizations (Stats)

Postby  philipp94831 » Tue Nov 28, 2017 9:11 pm

  28 Nov 2017, 20:07 GMT » Rey_Fer wrote:
  28 Nov 2017, 19:46 GMT » philipp94831 wrote:
  28 Nov 2017, 19:08 GMT » Rey_Fer wrote:
Could you explain more extently how does your agorithm work and what data you took into consideration?


The idea is to construct a graph where each civ is a node. The nodes are connected via directed edges. There is one edge between two civs A and B if A lost to B. The number of losses determines the weight of an edge. We normalize the weights such that the sum of weights of all outgoing edges is 1 for each node.
After we initialized the graph, we want to model the probability to reach a node if we randomly walk the graph. We start at a random node and then either visit another random node or visit one of its neighbors. The probability to visit any node randomly is uniformly distributed. The probability to visit one of the neighbors depends on the weight of the edge between them. From that, we can derive the probability to randomly visit a node which is equivalent to the probability that a random civ we lost to is X.

More information: https://en.wikipedia.org/wiki/PageRank


Do you have enough data in this tournament to make that graphe?


Hard to judge^^ I do think so because the ranking is reasonable but it remains to be seen whether the ranking is volatile or will remain quite stable
 philipp94831
 
Posts: 20
Joined: Mar 23, 2017
 

Re: [KotD] Clash of the Civilizations (Stats)

Postby  green eggs » Tue Nov 28, 2017 10:24 pm

  28 Nov 2017, 19:17 GMT » Rey_Fer wrote:
  28 Nov 2017, 13:43 GMT » misafeco wrote:
Higher ELO civ won 63 out of 94 times, which is roughly 2/3 of the games. That's obviously not very accurate. I don't think you can make an objective list even with hundreds of matches between each one. There are some lower rated civs which can counter a strong civ pretty well, the strenght of a civilization is more about dominating as many civs as possible. You can safely say that Aztecs, Incas, Indians etc. are top tier civs on Arabia and Ethiopians, Vietnamese and Malay are not and still have matches where the latter beats the former.
And of course balance is not just about Arabia. I wonder how easily Burmese dominates Arena with relics visible, cheap monk techs, Arambai and OP infantry while still being competitive in open maps like Arabia. We don't need perfect Arabia balance, where Teutons are just as good as Huns. But balance is obviously wrong if Huns beat Teutons on every single popular map.


I agree100% with you, In the previous version of AoC, Mayans was clearly a top3 civ for almost every map (including 4v4) ando probably considered by many as top1. Nevertheless, it was kind of countered by the Goths which was hardly ever picked in any free pick tournament.

...

So I think's it doesn't matter if some civs seem more powerful than others in arabia settings as in other particular settings some civ may be useful- What we cannot tolerate is a bunch of too OP civs and it feels sad if all new civs are sistematicly superior to classical civs (which already happened with AoC respect to AoK, with the korean exception although korea was also strong in BF maps).

So, let's take the numbers and then we valorate


These are good points, although Elo ratings are not meant to be definitive and illustrate only general strength levels. If an otherwise extremely powerful civ has a particular Achilles' heel (eg Mayans vs Goths), it won't affect the rating too much. I'm still inclined to think that a higher sample size will iron out many of these issues, but maybe tournament play is the wrong place to be drawing this particular sample since we will struggle to find enough data for civs that have fallen out of favour. (When are we ever likely to see the Saracens, for instance, let alone in statistically significant numbers?)
 green eggs
 
Posts: 21
Joined: Jul 28, 2017
 
Advertisement from Google 
 

Re: [KotD] Clash of the Civilizations (Stats)

Postby  JoshuaR » Wed Nov 29, 2017 2:08 am

I agree it would be amazing to carry forward the data from this tournament for future tournaments on at least standard land maps. We could see the civ choice meta evolve over time as the data becomes clearer and learn how boneheaded we all were way back when (now).

Though yes of course we could also just examine all rated games played over the past x months on WK and compare that data with the tourney data, as well.
 JoshuaR
 
Posts: 138
Joined: Oct 11, 2013
 

Re: [KotD] Clash of the Civilizations (Stats)

Postby  misafeco » Mon Dec 04, 2017 12:50 pm

New ELO standings after Quarter Finals:
1. Aztecs 1692
2. Incas 1679
3. Burmese 1652
4. Huns 1648
5. Mayans 1627
6. Britons 1626
7. Indians 1622
8. Goths 1599
9. Celts 1594
10. Mongols 1590
11. Malians 1582
12. Italians 1567
13. Berbers 1564
14. Magyars 1539
15. Vietnamese 1532
16. Ethiopians 1529
17. Malay 1494

Just a few observations:
-Meso civilizations won 5/6 with the only defeat is against another meso civ.
-Burmese still very strong with 1 win and 2 bans.
-Huns are starting to be very popular, picked all 4 times, 2 wins, 1 losses.
-Malians lost over 100 ELO points in their last 5 matches (5 losses).
-Goths and Italians are on the list now with 2 games each
 misafeco
 
Posts: 97
Joined: Feb 17, 2012
 1572 (42%)
 

Re: [KotD] Clash of the Civilizations (Stats)

Postby  philipp94831 » Mon Dec 04, 2017 1:08 pm

  04 Dec 2017, 11:50 GMT » misafeco wrote:
New ELO standings after Quarter Finals:
1. Aztecs 1692
2. Incas 1679
3. Burmese 1652
4. Huns 1648
5. Mayans 1627
6. Britons 1626
7. Indians 1622
8. Goths 1599
9. Celts 1594
10. Mongols 1590
11. Malians 1582
12. Italians 1567
13. Berbers 1564
14. Magyars 1539
15. Vietnamese 1532
16. Ethiopians 1529
17. Malay 1494

Just a few observations:
-Meso civilizations won 5/6 with the only defeat is against another meso civ.
-Burmese still very strong with 1 win and 2 bans.
-Huns are starting to be very popular, picked all 4 times, 2 wins, 1 losses.
-Malians lost over 100 ELO points in their last 5 matches (5 losses).
-Goths and Italians are on the list now with 2 games each


I'm sorry if I missed it but in which order do you compute ELO?
 philipp94831
 
Posts: 20
Joined: Mar 23, 2017
 

Re: [KotD] Clash of the Civilizations (Stats)

Postby  misafeco » Mon Dec 04, 2017 1:17 pm

  04 Dec 2017, 12:08 GMT » philipp94831 wrote:
I'm sorry if I missed it but in which order do you compute ELO?

First matches first. In the Quarter Finals it's Slam vs. TheMax -> Lierrey vs. ACCM -> TheViper vs. Vinchester -> DauT vs. Hearttt
 misafeco
 
Posts: 97
Joined: Feb 17, 2012
 1572 (42%)
 

Re: [KotD] Clash of the Civilizations (Stats)

Postby  Rorarimbo » Mon Dec 04, 2017 5:06 pm

Added the ELO update in the bifg file. Everything is updated there I think.
 Rorarimbo
 
Posts: 310
Joined: Jan 10, 2016
Location: Bulgaria
Age: 35
 
 
PreviousNext

Return to General

Who is online

Users browsing this forum: No registered users and 1 guest

Legend: Global moderators, News posters, Tournament moderators

cron