Page 1 of 1

ELO and DCSS

PostPosted: Thursday, 8th January 2015, 00:57
by triorph
Just for self-interest I wondered what would happen if you gave each character combo and each player an Elo score. The idea being that matching Elo scores means a player and a character combination should have a 50% win-rate . Using the backlog of online games I was able to achieve this. Note: there are assumptions made in the code that are probably not 100% fair. This only gives games of 0.14+. A score of 1200 is average for a player >= 5 wins in this period of time. Some of these results are skewed by competitions. I make no claims that any of these numbers really identify the easiest/hardest/best of anything, but thought that the numbers might be of interest for people to look at.

The results:

Player top 20:
1. keymashgrqeeg: 1888.34, 2. Tedronai: 1886.78, 3. bmfx: 1829.53, 4. Basil: 1805.93, 5. Sphara: 1794.31, 6. 78291: 1773.53, 7. zzxc: 1759.10, 8. ackack: 1754.39, 9. Zooty: 1729.28, 10. blakmane: 1699.36, 11. elliptic: 1677.08, 12. johnnyzero: 1666.31, 13. Lasty: 1658.35, 14. ManMan: 1628.07, 15. TheNoid: 1619.94, 16. ldf: 1611.21, 17. FiftyNine: 1582.12, 18. simm: 1580.70, 19. Berder: 1573.43, 20. Shadowmage952: 1572.73

Character combination easiest 20:
1. GrFi: 1293.11, 2. MfAs: 1358.95, 3. NaFi: 1425.71, 4. CeFi: 1425.75, 5. DDHe: 1469.77, 6. MiWn: 1470.38, 7. MiBe: 1477.72, 8. DDBe: 1477.82, 9. GrEE: 1481.55, 10. DDEE: 1483.40, 11. HOGl: 1494.69, 12. TrWr: 1501.90, 13. MiFi: 1505.61, 14. DDFi: 1510.39, 15. DDNe: 1520.45, 16. FeBe: 1525.96, 17. HOBe: 1530.80, 18. SpBe: 1532.88, 19. MiGl: 1534.37, 20. HaBe: 1535.95

Character combination hardest 20:
1. OpTm: 2009.50, 2. DjFE: 1957.91, 3. MuMo: 1948.20, 4. FoWr: 1930.98, 5. MuCK: 1929.06, 6. DgTm: 1927.42, 7. MuSk: 1926.26, 8. HESk: 1915.96, 9. SpAK: 1906.63, 10. TeSu: 1906.35, 11. OpSk: 1898.55, 12. FoWn: 1895.35, 13. HEAM: 1888.74, 14. MuNe: 1886.05, 15. OpSu: 1884.06, 16. TeFE: 1878.62, 17. OpWn: 1875.30, 18. DESu: 1875.01, 19. DEWr: 1874.56, 20. DEWn: 1870.65

Re: ELO and DCSS

PostPosted: Thursday, 8th January 2015, 01:07
by crate
It's "Elo", not "ELO". It's a guy's name.

Secondly I'm curious how your #1 player has a 4% winrate. That doesn't seem to make any sense to me based on the rest of what you're describing.

Re: ELO and DCSS

PostPosted: Thursday, 8th January 2015, 01:36
by Kramin42
He has a 4% winrate overall, but:
  Code:
!won keymashgrqeeg cv>=0.15
keymashgrqeeg (cv>=0.15) has won 82 times in 221 games (37.10%)
!streak keymashgrqeeg
keymashgrqeeg has 11 consecutive wins (MfFE, HEAK, SpHu, DgWr, FoHu, HEHu, HuWr, VSGl, FeAK, FoEn, TeIE; MfFi, MfGl, MfMo, MfHu, MfAs, MfAr, MfWn, MfBe, MfAK, MfCK, MfHe) and has won their last 9 games (HaHe, DgAE, TrHu, GrMo, CeHu, HaSu, GrMo, SpSu, CeEn).

he's doing great recently which has upped his score.

Re: ELO and DCSS

PostPosted: Thursday, 8th January 2015, 02:04
by triorph
Its also worth nothing that the k-factor used in this implementation is 32, which is definitely on the high side. This means it strongly favours recent games as an indication of true-skill rather than luck.

Re: ELO and DCSS

PostPosted: Thursday, 8th January 2015, 05:15
by chequers
Can you rank races and backgrounds independently by difficulty?

Re: ELO and DCSS

PostPosted: Thursday, 8th January 2015, 05:19
by Arrhythmia
crate wrote:It's "Elo", not "ELO". It's a guy's name.


Legitimately came in to this thread expecting a discussion of Electric Light Orchestra.

Re: ELO and DCSS

PostPosted: Saturday, 31st January 2015, 01:49
by Berder
I have one question: where did you get the data? Or if anyone else knows, where would be the right place to get the data for this kind of thing? Is it possible to download Sequell's game database?

Re: ELO and DCSS

PostPosted: Saturday, 31st January 2015, 02:16
by Sar
SpAK is really hard

wait

Re: ELO and DCSS

PostPosted: Saturday, 31st January 2015, 02:48
by Berder
Sar wrote:SpAK is really hard

wait

It doesn't seem like it should be hard, but
  Code:
<halberd> !lg * spak !boring / won
<Sequell> 18/4957 games for * (spak !boring): N=18/4957 (0.36%)
<halberd> !lg greatplayers spak !boring / won
<Sequell> 5/378 games for greatplayers (spak !boring): N=5/378 (1.32%)

Perhaps just bad luck, or possibly people attempting speedruns who usually don't speedrun. (someone who speedruns all the time would not bias the Elo of SpAK much because their player Elo would be low).

What explanation would you suggest?

Re: ELO and DCSS

PostPosted: Saturday, 31st January 2015, 02:51
by Sandman25
Berder wrote:What explanation would you suggest?


Speedrunners wanted to get first rune before XL 7 and never left Abyss?

Re: ELO and DCSS

PostPosted: Saturday, 31st January 2015, 02:51
by Sar
...people starving themselves with Banish/Corrupt? I dunno.

Edit: @Sandman25: you can't get a rune in starting Abyss but not everyone knows that, so I kinda like your explanation more. Also, I imagine people doing things like making a SpAK with spears and trying to kite some aboms for "easy" XP.

Re: ELO and DCSS

PostPosted: Saturday, 31st January 2015, 02:57
by Sandman25
Sar wrote:Edit: @Sandman25: you can't get a rune in starting Abyss but not everyone knows that, so I kinda like your explanation more. Also, I imagine people doing things like making a SpAK with spears and trying to kite some aboms for "easy" XP.


Thanks, I didn't know you cannot get the rune this way. I hope it is described somewhere. AK starts with
  Code:
Lugonu says: Depart the Abyss and spread corruption, my servant!
(use the ability menu to exit the Abyss)


Let's make it
  Code:
Lugonu says: Depart the Abyss and spread corruption, my servant!
(use the ability menu to exit the Abyss)
Abyssal rune cannot be found before you exit the Abyss.

Re: ELO and DCSS

PostPosted: Saturday, 31st January 2015, 03:01
by Sar
It's not just that the rune can't be found, you also get D:1 loot. Hm, actually, since rune is A:3 and deeper - does starting Abyss even spawn downstairs?

Special-casing it made sense when Abyss levels didn't exist (since one could startscum AKs until he got one with a rune in sight) but now that they do, I'm not sure it's worth the spoilers.

Re: ELO and DCSS

PostPosted: Saturday, 31st January 2015, 17:29
by Sprucery
It would be pretty strange imho if starting Abyss didn't have portals deeper, and the rune on A:3 - 5. I mean, what would be the reason for that?

Re: ELO and DCSS

PostPosted: Saturday, 31st January 2015, 18:17
by WalkerBoh
This is all irrelevant to player Elo but regardless... In the old days there was a chance of finding abyssal rune on any given tile, so AK abyss start was special cased so you couldn't start scum until getting a lucky abyssal rune laying around early on. I imagine a lot of SpAK's were killed this way.

In current crawl, the abyssal rune doesn't (or maybe only very rarely) appear outside of vaults on A:3-A:5 so I don't think there's much need for special casing stairs to not appear any more.

Re: ELO and DCSS

PostPosted: Saturday, 31st January 2015, 23:56
by Berder
The most luck I've ever had in pre-dungeon abyss: Once I killed an eye of draining and exited the abyss at XL 4.

Re: ELO and DCSS

PostPosted: Monday, 2nd February 2015, 00:08
by triorph
Berder wrote:I have one question: where did you get the data? Or if anyone else knows, where would be the right place to get the data for this kind of thing? Is it possible to download Sequell's game database?


Whoops I haven't been keeping track of this thread.

Yes its possible to download the game databases. If anyone is interested in the code (be warned its not great, especially the database fetching part) I have it available on github.

https://github.com/triorph-emacs/dcss-elo-analyser