Win probability statistics
- Updated Monday, May 23rd 2016 @ 16:56:46
Now that the top 24 bots are known, it's time to start discussing who is going to win BlockBattle.
Like in previous competitions, I'd like to share win probabilities with you. I estimated them based on historic game results of the last few months, the semi-finals schema, and random finals schemas.
In the table below, the first column is the ranking, based on expected amount in EUR.
The next column is the probability to make it to the finals (last 8). The column after that is the probability to become first in the finals.
The last column is the expected amount in EURO that the bot earns.Rank Bot P(finals) P(win) E(EUR) 1 hogeris 100 % 41 % 569 2 doppotra 90 % 27 % 415 3 Hohol 91 % 11 % 278 4 TaroKong 91 % 9 % 228 5 Kubisme 85 % 7 % 192 6 BBKing 54 % 3 % 106 7 ShapeShifter 69 % 1 % 66 8 kuburis 81 % 0 % 63 9 August6 51 % 1 % 60 10 dizzy 34 % 1 % 27 11 WildCat 18 % 0 % 19 12 Owl_Block 10 % 0 % 9 13 Real_Block 17 % 0 % 6 14 Atlantis 6 % 0 % 1 15 Aleks_Block 1 % 0 % 0 16 MightyBow 0 % 0 % 0 17 SunBlock 0 % 0 % 0 18 wow_Block 0 % 0 % 0 19 aztyur 0 % 0 % 0 20 tvv766 0 % 0 % 0 21 meduza 0 % 0 % 0 22 BBX 0 % 0 % 0 23 AldaBlock 0 % 0 % 0 24 GTAgency2 0 % 0 % 0 All probabilities are based on:
- The real semi-finals schema, with best to 10, with difference of at least 2
- Random finals schemas, with best to 6
- Estimated win probabilities per matchThe win probabilities of a specific match of bot1 against bot2 are based on the win probability per game of bot1 against bot2, which is estimated as the number of won games divided by the number of played games, from the following games:
- All games of the finals versions of bot1 against bot2
- If there are too few games (less than 20) of the finals versions of both bots, then I added the most recent game results (only games on or after May, 1st) of other versions of one or both of the bots, until a maximum of 20 games
- If there are still too few games (less than 20), then I estimated the remaining win probability based on the ELO's of both bots from the leaderboard of 15-05-2016, just before lockdownAre these probabilities statistically relevant? No, they are not, because the variance of the win probabilities is still quite high with only 20 games or even less between two bots. But the experience from the Warlight-2 and Texas Hold'em competitions learned that the ranks are pretty reliable.
I'll try to keep you updated after each semi-finals round.
- Created Monday, May 23rd 2016 @ 19:16:23
@Ads, are you a statistician?
Interesting data to have, thanks for your contribution to all this.
- Created Monday, May 23rd 2016 @ 19:46:15
No, I'm not a statistician. Professionally I'm a software engineer, but I do like to turn data into information.
I'm also composing some nice statistics about all bot's behavior. I'll share that here as well as soon as it's finished.
- Updated Monday, May 23rd 2016 @ 21:26:37
Hurray! I always look forward to your stats. Excited to see who will come out on top and if any surprises will happen. For example, I think meduza uploaded a new version right before lockdown so his/her chances might be a lot better than the statistics suggest.
- Created Tuesday, May 24th 2016 @ 00:25:32
Update after round 1:
No real surprises so far. After winning from Hohol, doppotra is now just behind hogeris. 4 bots are after round 1 already certain of a place in the top 8.
Rank Bot P(finals) P(win) E(EUR) 1 hogeris 100 % 34 % 514 2 doppotra 100 % 34 % 497 3 TaroKong 100 % 12 % 276 4 Kubisme 100 % 7 % 217 5 Hohol 78 % 6 % 192 6 BBKing 82 % 5 % 162 7 kuburis 58 % 0 % 43 8 dizzy 50 % 1 % 41 9 ShapeShifter 34 % 0 % 35 10 Owl_Block 24 % 0 % 19 11 Real_Block 50 % 0 % 18 12 August6 13 % 0 % 14 13 WildCat 12 % 0 % 11 14 MightyBow 1 % 0 % 0 15 aztyur 0 % 0 % 0 16 wow_Block 0 % 0 % 0 17 Aleks_Block 0 % 0 % 0 18 SunBlock 0 % 0 % 0 19 tvv766 0 % 0 % 0 20 meduza 0 % 0 % 0 21 Atlantis 0 % 0 % 0 22 BBX 0 % 0 % 0 23 AldaBlock 0 % 0 % 0 24 GTAgency2 0 % 0 % 0 - Created Tuesday, May 24th 2016 @ 13:25:09
Well, It was a close first round for Kubisme.A 5-0 lead ended up in (just) a 11-9. I expect ShapeShifter will make it to the final 8 anyway.
- Created Tuesday, May 24th 2016 @ 15:45:18
@Corniel Haha it was quite close. Good games! Many strong bots are already in the losers' bracket; I hope I can make it!
- Created Tuesday, May 24th 2016 @ 22:40:39
@cowzow: Yep, and you had some great improvements lately. We will see what comes out of that.
- Updated Wednesday, May 25th 2016 @ 00:10:26
Update after loosers round 2.
Hardly any changes compared to yesterday, because this time there were really no surprises.
Rank Bot P(finals) P(win) E(EUR) 1 hogeris 100 % 35 % 523 2 doppotra 100 % 33 % 484 3 TaroKong 100 % 12 % 278 4 Kubisme 100 % 7 % 215 5 Hohol 79 % 6 % 197 6 BBKing 81 % 5 % 160 7 dizzy 51 % 1 % 45 8 kuburis 58 % 0 % 44 9 ShapeShifter 32 % 0 % 32 10 Real_Block 50 % 0 % 18 11 Owl_Block 24 % 0 % 18 12 August6 13 % 0 % 15 13 WildCat 11 % 0 % 12 14 MightyBow 0 % 0 % 0 15 aztyur 0 % 0 % 0 16 Aleks_Block 0 % 0 % 0 17 wow_Block 0 % 0 % 0 18 meduza 0 % 0 % 0 - Created Thursday, May 26th 2016 @ 00:17:42
After winners round 3:
The six pool winners are known: hogeris, BBKing, TaroKong, artoppod, kuburis and kubisme.
Most likely candidates for the remaining places in the top 8 are Hohol and ShapeShifter.
Rank Bot P(finals) P(win) E(EUR) 1 hogeris 100 % 38 % 543 2 artoppod 100 % 30 % 462 3 TaroKong 100 % 10 % 249 4 Hohol 80 % 9 % 228 5 Kubisme 100 % 7 % 201 6 BBKing 100 % 5 % 195 7 kuburis 100 % 0 % 63 8 ShapeShifter 55 % 1 % 46 9 Owl_Block 31 % 0 % 22 10 August6 18 % 0 % 17 11 WildCat 12 % 0 % 12 12 Real_Block 2 % 0 % 1 13 dizzy 2 % 0 % 1 14 aztyur 0 % 0 % 0 15 meduza 0 % 0 % 0 16 Aleks_Block 0 % 0 % 0 17 MightyBow 0 % 0 % 0 18 wow_Block 0 % 0 % 0 - Updated Thursday, May 26th 2016 @ 01:14:21
Isn't my P(finals)=80% way too optimistic? WildCat is quite strong...
And I've just realized that I don't understand, what's going on in round 8. Three participants, two winners? - Updated Thursday, May 26th 2016 @ 08:13:50
"All bots play against each other once."
Really? Exactly once? Not "one series of matches"?
"So then the bot that used the most rounds to win his games is out of the tournament"
What? In-game number of rounds? SUDDENLY it is important? Not just win/loss condition?If I understand it right, these rules sound plain stupid to me.
"We used bunch of Bo10 series to get best bots of loser bracket. Now let's kick one of them randomly, LOL" - Created Thursday, May 26th 2016 @ 09:11:59
@Hohol: no, it's not like that. In the last round still Bo10 matches are played between the three remaining bots. If one of the three looses both of his matches, then he is out. Only in case each of the three competitors wins one match and looses the other, then the number of games in these matches determines which two continue to the finals. That case is quite exceptional. As far as I remember, in all previous competitions, this never happened.
@Hohol: about your 80% probability to proceed to the last 8: according to my statistics, you will win your next match against MightyBow with 100% probability (per game 80%) and the next one against WildCat with 80% probability (per game 59%). In the last round, you are as good as certain to be one of the best two.
