<< Back to Warzone Classic Forum   Search

Posts 1 - 14 of 14   
1v1 Ladder Statistics: 4/25/2020 15:34:44


JustinR17 
Level 59
Report
Hello everyone! Since the end of December, I spent a lot of time into gathering games from every ladder and (so far) analyzing the 1v1 ladder games. I'll share some summarized data in this thread, although most of the data can be found in full on a sheets I created:
https://docs.google.com/spreadsheets/d/1wQrylkSl0QTKXT4haQp_Q47OROYVrqLmVHAfJk1EeDQ/edit?usp=sharing

Using the WZ API, I collected every ladder game and analyzed everything using various scripts. I'd like to thank Krunx for his massive help around rating systems, analyses and data collection in general. I'd also like to thank Derfellios for coming up with a ton of different analyses to look into, AI similarly with overall help and Knyte for helping me get started on this project in the first place. Lastly, thank you Fizzer for even allowing this to be possible with creating the API and for putting up with my questions :)

(** All data is as of December 24 2019 **) -- Some of the stats below isn't too friendly on mobile (because of unity i think?). Use the web version.

As of December 24 2019, there have been a total of 292 777 1v1 ladder games (both old/new ladder) while the old ladder accounts for 123 428 games (~42% of all 1v1 ladder games). Since the creation of the ladder, there have been 3 832 players and only ~700 players have played fewer than 20 games. Of the total players, 2 091 have played on the old ladder and 2 551 have played on the new ladder (~54.6% and ~66.6% of total players respectively -- 810 played on both versions). There have been 32 846 games (~11% of all 1v1 ladder games) that ended on picks and 2 388 games ending immediately after picks. The total number of boots on the ladder is 52 400.


In order to download all of the data, I had to do it in rounds on my RaspberryPi due to running out of space and internet drops. In total, the raw API response for all 1v1 ladder games took up 43.3GB. To not kill my laptop nor to wait hours to run scripts, I split the data into 3 databases. The first database contained basic game info like who won/lost/number of turns/end date/player data and ratings. This took up only 155.1MB of space. The second database contained all orders for every game and was split into individual turns. This took up 7.3 GB of space. The last database contained all of the map standings for every game and was split into individual turns as well. This took up 19.9GB of space.

Overall, the 1v1 ladder games accounted for about 60% of the total space for all ladders (including seasonals + RTL). With regards to code, there are a total of 5 254 lines for the project with ~4 000 lines for analysis and ~1 000 lines for downloading the data and setting up the databases.


The average turn length of all games is 10.32 turns (std dev: 6.51 turns). If you exclude games that end on picks, the average is 11.63 turns (std dev: 5.70 turns). Excluding games that are 0 turns long (ending before or after picks), the average length is 11.73 turns (std dev: 5.61 turns).

Distribution of game lengths (Excluding games ending on picks):

(image link if didn't load above: https://imgur.com/clTXMfb)

(Sorry, my statistics knowledge is lacking) The distribution of game lengths appear to be roughly normal. There does appear to be a slight skew to the right which would account for the outliers. I would do more analysis, but I have other work to do :P

Assuming for now the fact that the data follows a normal distribution and game length is independent/random, the probability of a 152 turn game occuring by random chance is ~1.789x10^-138.

Top 10 longest games by turns:
1. 152 Turns; Kurulananfok vs Carlos Danger; https://www.warzone.com/MultiPlayer?GameID=6740796 (old ladder) 
2. 147 Turns; DobbieDice vs N-W-G; https://www.warzone.com/MultiPlayer?GameID=5278322 (old ladder) 
3.  85 Turns; CJ vs Legacy; https://www.warzone.com/MultiPlayer?GameID=9885551 (new ladder) 
4.  59 Turns; Kris[MD] vs {101st}Blowfly; https://www.warzone.com/MultiPlayer?GameID=5922011 (old ladder) 
    59 Turns; Cancer vs raftibaba; https://www.warzone.com/MultiPlayer?GameID=7883570 (old ladder) 
6.  58 Turns; Jayvan vs Leo41; https://www.warzone.com/MultiPlayer?GameID=7980000 (old ladder) 
    58 Turns; {101st} Grickface vs Itphoto; https://www.warzone.com/MultiPlayer?GameID=11895995 (new ladder) 
8.  57 Turns; Janine vs Elan; https://www.warzone.com/MultiPlayer?GameID=13745292 (new ladder) 
9.  56 Turns; [BR]DuckKing vs raftibaba; https://www.warzone.com/MultiPlayer?GameID=7918889 (old ladder) 
    56 Turns; Prabster Realm vs vogelpu; https://www.warzone.com/MultiPlayer?GameID=10673769 (new ladder) 


That first game on the list should have barely made it past 20 turns, but was possibly one of the most satisfying games with watching Carlos go from 97-5 inc and a stack of 4 300 to 5-96 inc with a 6 000 stack, only to sweep back through the board to win.

Second game was similar with the losing DobbieDice begging to win in order to get 10 ladder wins in a row. Of course a few turns later, it seemed to be the perfect opportunity to ask the opponent to join their clan (didn't seem to work :'( ). At the end, it lasted 140 more turns than necessary and with DobbieDice winning due to a boot despite being at a 96 income deficit and a ~12 000 troop deficit.

Congrats to both Kris[MD]/{101st}Blowfly and Cancer/raftibaba for getting the longest legit 1v1 ladder games. Wise words of {101st}Blowfly in his game, "he who blinks first is lost!!!!"


Top 10 longest games by time:
1. 109 days; Heyheuhei vs Troll;  https://www.warzone.com/MultiPlayer?GameID=1600721 (old ladder) 
2.  94 days; majokun vs Artctic82; https://www.warzone.com/MultiPlayer?GameID=7883264 (old ladder) 
3.  86 days; Fizzer vs Troll; https://www.warzone.com/MultiPlayer?GameID=1720440 (old ladder) 
4.  86 days; Robert E Lee vs ACL Tears; https://www.warzone.com/MultiPlayer?GameID=6739530 (old ladder) 
5.  84 days; Gilgamesh vs {101st}Blowfly; https://www.warzone.com/MultiPlayer?GameID=16763131 (new ladder) 
6.  83 days; El professor vs Coronel Gavilan; https://www.warzone.com/MultiPlayer?GameID=10490898 (new ladder) 
7.  77 days; ljgmsun vs Vulpes; https://www.warzone.com/MultiPlayer?GameID=5553073 (old ladder) 
8.  77 days; Bbraw vs Ouy Gnillats; https://www.warzone.com/MultiPlayer?GameID=17061222 (new ladder) 
9.  75 days; TheWarlightMaster vs Vulpes; https://www.warzone.com/MultiPlayer?GameID=5566865 (old ladder) 
10. 73 days; FiveStarGeneral vs Quicksilver; https://www.warzone.com/MultiPlayer?GameID=15895404 (old ladder) 


Suprisingly, many of these games didn't go into too many turns. Troll was the first one on the list and third boasting impressive 10 day and 14 day turn speeds respectively.


Top 10 players by win count:
1.  1660 wins - Bigchps147 (2915 games - 0.570 win rate)
2.  1614 wins - adiba (3107 games - 0.520 win rate)
3.  1552 wins - Anakin Chigurh (2729 games - 0.569 win rate)
4.  1439 wins - Klefaz (2743 games - 0.525 win rate)
5.  1319 wins - Sebus Maximus (2533 games - 0.521 win rate)
6.  1230 wins - konglaide (2432 games - 0.506 win rate)
7.  1201 wins - Legend of the Phoenix (2365 games - 0.508 win rate)
8.  1076 wins - jasdanmoo(2199 games - 0.489 win rate)
9.  1031 wins - JaponeSafado (1954 games - 0.528 win rate)
10. 1020 wins - Robert E Lee (2001 games - 0.510 win rate)


Top 10 players by game count:
1.  3107 games - adiba
2.  2915 games - Bigchps147
3.  2743 games - Klefaz
4.  2729 games - Anakin Chigurh
5.  2533 games - Sebus Maximus
6.  2451 games - {101st}Blowfly
7.  2432 games - konglaide
8.  2365 games - Legend of the Phoenix
9.  2199 games - jasdanmoo
10. 2001 games - Robert E Lee


Top 10 players by win rate (must have >= 100 games):
1.  0.8055 win rate - Kurdistan49
2.  0.7873 win rate - alhazi
3.  0.7818 win rate - Buns157
4.  0.7785 win rate - Equivoque
5.  0.7771 win rate - AlturoSensei
6.  0.7692 win rate - The Impaller
7.  0.7545 win rate - MisterT
8.  0.7410 win rate - TeddyFSB
9.  0.7375 win rate - Guiguzi
10. 0.7368 win rate - FiveStarGeneral


Edited 4/26/2020 03:28:53
1v1 Ladder Statistics: 4/25/2020 15:35:08


JustinR17 
Level 59
Report
For a bit of fun, I also looked at boot statistics. Over the 292 777 1v1 ladder games, there have been 52 400 boots (~17.9% of all games). On a better note, there are 1 287 players (~33.6% of all players) that have 0 boots. Looking at the top players by boot rate (w/ >= 20 games), there was one person with a perfect 20 boots out of 20 games (congrats?).

Top 10 players by boots:
1.  617 boots - NZPhoenix (AHOL) (1532 games - 0.403 boot rate)
2.  505 boots - swat (1580 games - 0.320 boot rate)
3.  383 boots - Wiredawg (1656 games - 0.231 boot rate)
4.  373 boots - BarnsleyVeteran (490 games - 0.761 boot rate)
5.  327 boots - Kenraali (752 games - 0.435 boot rate)
6.  313 boots - DeathBecomesYou (465 games - 0.673 boot rate)
7.  280 boots - Drackoe (735 games - 0.381 boot rate)
8.  269 boots - ajmayo (1118 games - 0.241 boot rate)
9.  266 boots - The Girl (535 games - 0.500 boot rate)
10. 259 boots - Alexandre Cassagne (577 games - 0.449 boot rate)



I also applied various rating systems throughout all ladder games (Glicko2 w/ 1d rating period and Elo w/ k=32)

Top 10 players for all 1v1 ladder games under Glicko2:
1.  2270 - 89thlap (Std. Dev. 152)
2.  2262 - Rufus (Std. Dev. 128)
3.  2232 - Marcin (Std. Dev. 94)
4.  2225 - Krzysztof (Std. Dev. 78)
5.  2225 - NoMercy (Std. Dev. 279)
6.  2214 - Kurdistan49 (Std. Dev. 57)
7.  2198 - #Master [QB] (Std. Dev. 314)
8.  2184 - Master Moto - モトキ (Std. Dev. 184)
9.  2178 - Buns157 (Std. Dev. 323)
10. 2172 - Latnox (Std. Dev. 437)


Top 10 players for all ladder games under Elo:
1.  2103 - Kurdistan49
2.  2075 - Krzysztof
3.  2035 - malakkan
4.  2028 - Edge
5.  2023 - Buns157
6.  2011 - AlturoSensei
7.  2003 - AI
8.  1988 - Leopard
9.  1980 - Latnox
10. 1976 - GiantFrog



Finally, I also did analysis on what the most common picks were. This included both which bonuses were most picked, and also which territories within a bonus got picked the most.

Bonuses sorted by most picked on the new ladder:
1.  Scandinavian Peninsula (9.04%)
2.  Central America (8.67%)
3.  Antarctica (7.76%)
4.  West Russia (7.54%)
5.  South America (7.15%)
6.  Southeast Asia (6.81%)
7.  Central Russia (6.73%)
8.  East China (4.92%)
9.  Indonesia (4.74%)
10. East US (4.70%)
...
21. Middle East (1.24%)
22. Europe (1.09%)
23. Canada (1.08%) :(


Bonuses Picked Distribution:

(image link: https://imgur.com/Fvxmf6Z)

Bonuses sorted by most picked on the old ladder:
1.  Scandinavian Peninsula (9.61%)
2.  Central America (8.66%)
3.  Antarctica (8.58%)
4.  West Russia (8.45%)
5.  South America (7.94%)
6.  Southeast Asia (7.79%)
7.  Central Russia (6.40%)
8.  East China (5.79%)
9.  West Africa (5.71%)
10. Indonesia (5.46%)
...
21. East US (0.98%)
22. East Russia (0.69%)
23. Canada (0.67%) :(


Bonuses Picked Distribution:

(image link: https://imgur.com/oPSPKmd)
This post is getting really long, so I won't go into the most picked territories in a bonus. This info is all in the sheets attached.

Once I get time again (~Sept.), I was thinking about looking at the other ladders. If anyone has suggestions on what to look at next, feel free to mention or pm me :)

Edited 4/25/2020 15:39:23
1v1 Ladder Statistics: 4/25/2020 15:55:42


AI 
Level 64
Report
Thank you very much Justin! Awesome work and many interesting insights.
1v1 Ladder Statistics: 4/25/2020 16:01:06


Aura Guardian 
Level 62
Report
Hi justin,

This is pretty cool!

Just curious, what file format were you saving your data in?

Also, what language did you use to run your analysis?

-Aura

Edited 4/25/2020 16:03:35
1v1 Ladder Statistics: 4/25/2020 16:30:00


JustinR17 
Level 59
Report
Hey Aura,

Everything I wrote was in javascript. I had downloaded the data originally using js and so decided to stick with it (and because of familiarity). There were a few stages before I was able to run any analysis. The first stage involved downloading all of the games and I ended up storing it as json files. IIRC that took about a week as the biggest bottleneck was the API and having to run a single request at a time. Once that was all done, I had to upload all of the data into databases. I opted to use MongoDB since it worked well with JS, although it took me a couple tries of uploading the entire game data because I was hitting the resource limits on my laptop for queries + timeouts.

This is why I went with 3 collections with varying data. The smallest collection in the db with general game info was the one most hit, while the orders and map standings were only used on later analyses.
1v1 Ladder Statistics: 4/25/2020 16:39:51


Phobos 
Level 62
Report
Great work Justin! I find the bonus pick frequencies and # of turns per game the most interesting. Also those super long games are hilarious.
1v1 Ladder Statistics: 4/25/2020 17:30:35


Viking1007 
Level 59
Report
Wow! Amazing work Justin!

The longest games are hilarious, as Glass said! Very funny to read the chats in them! The bootrate statistics and the most common picks are also interesting, seeing how many times Boston got booted lol.

Keep up the good work!
1v1 Ladder Statistics: 4/25/2020 17:54:55


DanWL 
Level 62
Report
Nice stats. Google way to comapre everyone's skill
1v1 Ladder Statistics: 4/26/2020 12:17:56


Math Wolf 
Level 63
Report
The distribution of the number of turns would be Poisson distributed rather than normal - a Poisson with high average is approximately normal, hence the similar shape.

Very interesting dataset, I'll take a look at it when I have a bit more time. :-)
1v1 Ladder Statistics: 4/28/2020 15:00:19

[V.I.W] recruiting time! Join us !
Level 63
Report
With that game Carlos is the king of weirdos.

great work man
1v1 Ladder Statistics: 4/28/2020 15:55:10


OvertForeigner 
Level 55
Report
Yes. Letting opponent conquer everything just to retake it with a giant stack was really funny.
1v1 Ladder Statistics: 4/28/2020 16:01:40


goralgn
Level 55
Report
all my wins were boots


JUST THE WAY I LIKE IT
1v1 Ladder Statistics: 4/29/2020 02:55:54


JustinR17 
Level 59
Report

The distribution of the number of turns would be Poisson distributed rather than normal - a Poisson with high average is approximately normal, hence the similar shape.

@MW, I knew something felt off while trying to do it. Just learned this a couple months ago and already forgot :P
1v1 Ladder Statistics: 4/29/2020 13:55:06


Automatix 
Level 60
Report
Awesome!
Posts 1 - 14 of 14