19 coaches online • Server time: 06:01
Index Search Usergroups Profile
Log in
Recent Forum Topics goto Post Crown of Sand IX Fin...goto Post Got a weird one - Th...goto Post Custom Ruleset apply...
SearchSearch 
Post new topic   Reply to topic
View previous topic Log in to check your private messages View next topic
Sp00keh



Joined: Dec 06, 2011

Post   Posted: Jun 13, 2018 - 14:01 Reply with quote Back to top

Page 5 has definitely got problems.. it's an attempt to condense / combine the separate charts from previous page of TD ratio and CAS ratio, along with winrate...
But yea I'm open to suggestions to improve it
Sp00keh



Joined: Dec 06, 2011

Post   Posted: Jun 13, 2018 - 14:22 Reply with quote Back to top

Ok maths question

Race1 wins 57%
Race2 wins 48%

If they play each other, what is the expected outcome of VS winrate? How do I work this out?

If race1 wins 60% of the games Vs race2, how do I know how much above or below the expected it is

I want to find racial advantages from skillset that have difference from expected
Eg dwarves and Amazon might both win a lot but do dwarves crush Amazons? What other matchups are there with big discrepancies?
bghandras



Joined: Feb 06, 2011

Post   Posted: Jun 13, 2018 - 15:21 Reply with quote Back to top

There are many methods to tackle it, and you will get some level of contradiction.(Which is fine, as there are other factors like matchup that provide deviation.) I would recommend a simulation using maximum likelyhood method, which would give you an expected win rate for every team. Then rerun it and see what is the correlation of its prediction vs reality. The rest then can be further analysed for matchup, coaching, etc.

_________________
Image
'Best thing on FUMBBL'(Seventyone),
'AWESOME'(Kummo), 'Last league i would leave'(SpecialOne), 'Far far the BEST thing i played online'(Grunth)
koadah



Joined: Mar 30, 2005

Post   Posted: Jun 13, 2018 - 15:48 Reply with quote Back to top

Sp00keh wrote:
Ok maths question

Race1 wins 57%
Race2 wins 48%

If they play each other, what is the expected outcome of VS winrate? How do I work this out?

If race1 wins 60% of the games Vs race2, how do I know how much above or below the expected it is

I want to find racial advantages from skillset that have difference from expected
Eg dwarves and Amazon might both win a lot but do dwarves crush Amazons? What other matchups are there with big discrepancies?


Is there enough data to consider TV? Dwarves vs chaos at 1000TV is not the same as at 2000.

Garion might be interested in a Secret League version. He was asking about data a week or so ago.

_________________
Image
Instant Massive teams - 8 team KO tournament - 7th June
Sp00keh



Joined: Dec 06, 2011

Post   Posted: Jun 13, 2018 - 16:05 Reply with quote Back to top

It starts to get fragmented
24 races * 24 opponents * TV divisions, say 5 splits = 2880 combos
That'd be an average of 25 games in each category, for the data I did this with originally

Thank you for newer data
At the same rate of games played it'll be nearly double that amount

Still probably shakey to draw much conclusions from.. eg orcs Vs chaos we'd have enough data at all the TV brackets
Goblins Vs slann at 1800tv = not enough games played Smile
Christy



Joined: Jul 19, 2007

Post   Posted: Jun 13, 2018 - 16:28 Reply with quote Back to top

A quick method would be A* (1-B)/.5

So in your example it would be 0.57 *(1-0.4Cool/0.5 = 0.5928 for team A to win.

Then you could also have 0.48*(1-0.57)/0.5 = 0.4128. As pointed out above you get a consistency issue (they don't sum to 1).
For some very extreme cases they may also not be in the range (0, 1) but these will also be fixed by the reweighting below.

The logic is weighting your own win percentage by the edge an opponent gives against an average opponent. If you win 70% of matches and your opponent wins 40%, they lose 20% more often than they should vs an average opponent (60% of the time instead of 50%). If they keep that up across the board they should lose to you 70% * 120% of the time.

You could also reweight as 0.5938/(0.5928+0.4128)=0.589499 and 0.4128/(0.5938+0.4128)=0.410501. Reweighting will likely only change much in the extreme cases though (lots of wins vs barely any wins). This ensures they sum to 1 and neither element is above 1.
koadah



Joined: Mar 30, 2005

Post   Posted: Jun 13, 2018 - 16:44 Reply with quote Back to top

You might want to check out this feller.

Try to forgive his bedside manner. Wink

_________________
Image
Instant Massive teams - 8 team KO tournament - 7th June
PurpleChest



Joined: Oct 25, 2003

Post   Posted: Jun 13, 2018 - 17:51 Reply with quote Back to top

I don't know whats possible, but for page 5 i would either:

Sort the races by win rate and present the same graph but with the races sorted that way 1-24. This would make one line a simple descent, but would show interestingly the other two in easier comparison.

Or go with bars, as the current race order doesnt justify linking the points.

_________________
Barbarus hic ego sum quia non intelligor ulli
I am a barbarian here because i am not understood by anyone
Zirgantz



Joined: May 30, 2017

Post   Posted: Jun 13, 2018 - 18:30 Reply with quote Back to top

Fantastic work !

I like how Ogres are first both in HomeCas and in AwayCas categories.
Sp00keh



Joined: Dec 06, 2011

Post   Posted: Jun 14, 2018 - 09:42 Reply with quote Back to top

Updated to yesterday, now has 152k games
Balle2000



Joined: Sep 25, 2008

Post   Posted: Jun 14, 2018 - 10:39 Reply with quote Back to top

I think you put in a lot of work in this. Well done, and thank you.
dabassman



Joined: Feb 17, 2006

Post   Posted: Jun 14, 2018 - 14:25 Reply with quote Back to top

Awesome dashboard, thank you!
Just one question (because I got slightly confused):
Shouldn't "W/L Ratio" just be called "WinRatio"?
Because, I suppose we are looking at the Win/Games Ratio (edit: games minus draws), not the Win/Loss Ratio...
But maybe this is just the common way to write it, in this case, ignore my question Very Happy
Sp00keh



Joined: Dec 06, 2011

Post   Posted: Jun 14, 2018 - 16:21 Reply with quote Back to top

It's the ratio between wins and losses
It ignores Draws, yes
So it's not wins per game. W/L ratio is then more correct, no?
Jopotzuki



Joined: Oct 07, 2011

Post   Posted: Jun 14, 2018 - 19:54 Reply with quote Back to top

Why no dwarves and Necros on page 2?

_________________
Kam wrote:
My Flings also disappear when they're blocked... I knew that was a bug!
ex-convict



Joined: Jun 28, 2005

Post   Posted: Jun 14, 2018 - 20:39 Reply with quote Back to top

No offense but this clearly needs to be taken down, otherwise the secret dominance of lizardmen will be realized by all.


Great work though.

_________________
Part of the NCBB and the NBFL.

Don't drop the soap.
Display posts from previous:     
 Jump to:   
All times are GMT + 1 Hour
Post new topic   Reply to topic
View previous topic Log in to check your private messages View next topic
Powered by PNphpBB2 1.2 © 2003 PNphpBB Group
Credits