EP Win Rate Statistics
Re: EP Win Rate Statistics
I built a table with jp elo @Goodspeed and also for NWC @BrookG
gotta say there the smurf problem might still have a big impact, with indiabot having a max jp elo of 1900 for instance...
PS : I did it with max 1v1 jp elo, not current one.
gotta say there the smurf problem might still have a big impact, with indiabot having a max jp elo of 1900 for instance...
PS : I did it with max 1v1 jp elo, not current one.
- Attachments
-
-
Re: EP Win Rate Statistics
The sample size in NWC is still small, but in the case of Brits for example it's 109 games which is not bad. The -72 average elo difference definitely explains the low win rate. Considering +100 ELO means 66% win rate (if the ELO algorithm was implemented correctly) Brit's real win rate would be more like 55%.
An interesting plot twist is that Iro's win rate on the other hand would actually turn negative, even though it's the highest win rate at first glance. Only 21 games though.
Do you think you could add a column that recalculates the win rates based on the average ELO differences? I don't have a formula for you, would have to think about it
An interesting plot twist is that Iro's win rate on the other hand would actually turn negative, even though it's the highest win rate at first glance. Only 21 games though.
Do you think you could add a column that recalculates the win rates based on the average ELO differences? I don't have a formula for you, would have to think about it
Re: EP Win Rate Statistics
Sadly any tournament sample is too small
Correlation doesn't mean causation.
http://www.tylervigen.com/spurious-correlations
"mr.brookg go buy jeans and goto the club with somppuli" - Princeofkabul, July 2018
http://www.tylervigen.com/spurious-correlations
"mr.brookg go buy jeans and goto the club with somppuli" - Princeofkabul, July 2018
Re: EP Win Rate Statistics
Yeah we still have not that many games on this datas. Although, Imo it's still better to have a look at this kind of stats than relying on pure feeling.
I found also interesting to have some data visualization (here for NWC). Here I did min-max normalization, that's why it's flat for iroquois. For me it's pretty clear on this image that china was the most powerful civ during NWC. Although as you said, only 44 games is not enought to make have this conclusion.
I found also interesting to have some data visualization (here for NWC). Here I did min-max normalization, that's why it's flat for iroquois. For me it's pretty clear on this image that china was the most powerful civ during NWC. Although as you said, only 44 games is not enought to make have this conclusion.
Re: EP Win Rate Statistics
Yeah 52% despite -67 ELO is pretty significant.
Why take max elo instead of current elo btw? Current should be more accurate no?
Why take max elo instead of current elo btw? Current should be more accurate no?
Re: EP Win Rate Statistics
Idk I though max was more accurate, especially since it's computed now and not at the time the game were played so current elo is current elo on the 18/10/2019 and not at the time the game were played.Goodspeed wrote:Yeah 52% despite -67 ELO is pretty significant.
Why take max elo instead of current elo btw? Current should be more accurate no?
Idk though, maybe the best would be to use what's currently used for seeding ? so max ep+current jp as I understood.
Re: EP Win Rate Statistics
Actually I think that the good thing to look at is expected winrate according to average elo difference versus real winrate. It gave me this for NWC :
- Attachments
-
Re: EP Win Rate Statistics
Should include something like % of possible match up covered by each civ games because imo its impossible to interpret if for exemple half of india game were india-russia. So atleast we know then on how many different mu each civ'winrate is based on.
Re: EP Win Rate Statistics
You would be decreasing even more the sampleflontier wrote:Should include something like % of possible match up covered by each civ games because imo its impossible to interpret if for exemple half of india game were india-russia. So atleast we know then on how many different mu each civ'winrate is based on.
Correlation doesn't mean causation.
http://www.tylervigen.com/spurious-correlations
"mr.brookg go buy jeans and goto the club with somppuli" - Princeofkabul, July 2018
http://www.tylervigen.com/spurious-correlations
"mr.brookg go buy jeans and goto the club with somppuli" - Princeofkabul, July 2018
Re: EP Win Rate Statistics
No i dont speak about decreasing, its just some more informations, if there has been 15 games for sioux, but you say they have play only 3 different mu's, you know that winrate is irrelevant and dont show anything about the strenght/balance of the civ.
Re: EP Win Rate Statistics
Nice. This graph is pretty meaningful I would say, except Brit expected winrate of 30% with -72 ELO can't be right. It would be 33.3% with -100. Must be something wrong with your formulabwinner wrote:Actually I think that the good thing to look at is expected winrate according to average elo difference versus real winrate. It gave me this for NWC :
Re: EP Win Rate Statistics
Hi, I am back with it, indeed there was an error, I divided by 200 instead of 400 somewhere.Goodspeed wrote:Nice. This graph is pretty meaningful I would say, except Brit expected winrate of 30% with -72 ELO can't be right. It would be 33.3% with -100. Must be something wrong with your formulabwinner wrote:Actually I think that the good thing to look at is expected winrate according to average elo difference versus real winrate. It gave me this for NWC :
There is the correct image
- Attachments
-
Re: EP Win Rate Statistics
Now all you need to do is sort by the delta But China top civ confirmed.
It's still too small of a sample I would say, but it does show that for example Iro's high win rate doesn't necessarily mean they're strong.
It's still too small of a sample I would say, but it does show that for example Iro's high win rate doesn't necessarily mean they're strong.
Re: EP Win Rate Statistics
What do you mean by sort by delta ?Goodspeed wrote:Now all you need to do is sort by the delta But China top civ confirmed.
It's still too small of a sample I would say, but it does show that for example Iro's high win rate doesn't necessarily mean they're strong.
Btw I am thinking also about using a regression to find what the best combination of current/max ep/jp elo is to predict who wins a match (this might be interesting for @Cometk as it could be use for seeding)
I think this analysis is a good tool to determine what civ has performed well or baddly during the tourney at least.
Re: EP Win Rate Statistics
Subtract expected win rate from win rate and sort by the resultWhat do you mean by sort by delta ?
-
- Gendarme
- Posts: 5996
- Joined: Jun 4, 2019
Re: EP Win Rate Statistics
@bwinner can you do this for the ep ELO data? @ESOCommunity please colaborate with bwinner to make this possible. I believe these are very powerful and meaningful statistics.
Also ideally we include the standard deviation for the civ winrates in the data. That'd be a very complete data analysis for civ balance in my eyes.
Also ideally we include the standard deviation for the civ winrates in the data. That'd be a very complete data analysis for civ balance in my eyes.
Re: EP Win Rate Statistics
Yes I can do that no problem, @Development Team help could be great to do the statistic on all match that happend on ep rather than only tourney (as they seem to have access to it). I personnally don't know where to find such datas.RefluxSemantic wrote:@bwinner can you do this for the ep ELO data? @ESOCommunity please colaborate with bwinner to make this possible. I believe these are very powerful and meaningful statistics.
Also ideally we include the standard deviation for the civ winrates in the data. That'd be a very complete data analysis for civ balance in my eyes.
Re: EP Win Rate Statistics
cant tell which are Germans and which are Ottomans from those pics
otherwise really nicely done, so China, Brit and one of the -mans op confirmed
otherwise really nicely done, so China, Brit and one of the -mans op confirmed
-
- Gendarme
- Posts: 5996
- Joined: Jun 4, 2019
Re: EP Win Rate Statistics
It claims Brits is underperforming right? Lower winrate than expected winrate based on ELO? So it's weakJerimuno wrote:cant tell which are Germans and which are Ottomans from those pics
otherwise really nicely done, so China, Brit and one of the -mans op confirmed
Re: EP Win Rate Statistics
Yeah not sure why it did that to the pic, anyway ger are the one who have performed wellJerimuno wrote:cant tell which are Germans and which are Ottomans from those pics
otherwise really nicely done, so China, Brit and one of the -mans op confirmed
Re: EP Win Rate Statistics
depends if you take ep elo or jp elo actually^^ That's why we need to find out what's the best measure, which I am working onRefluxSemantic wrote:It claims Brits is underperforming right? Lower winrate than expected winrate based on ELO? So it's weakJerimuno wrote:cant tell which are Germans and which are Ottomans from those pics
otherwise really nicely done, so China, Brit and one of the -mans op confirmed
Re: EP Win Rate Statistics
Brit has a higher win rate than expected according to above graphRefluxSemantic wrote:It claims Brits is underperforming right? Lower winrate than expected winrate based on ELO? So it's weakJerimuno wrote:cant tell which are Germans and which are Ottomans from those pics
otherwise really nicely done, so China, Brit and one of the -mans op confirmed
Re: EP Win Rate Statistics
hm, i was thinking the graph tells me brits have higher actual winrate than expected winrate? thats why i said brit op
yea what gs says
yea what gs says
-
- Gendarme
- Posts: 5996
- Joined: Jun 4, 2019
Re: EP Win Rate Statistics
Oh, I was looking at the EP win rate statistics? That made more sense to me.
Who is online
Users browsing this forum: No registered users and 20 guests