Statistics on Coders Strike Back Podium

#1

For anybody interested, here are a few statistics taken from some players’ match history.

I find it interesting to see for example :

  • Jeff loses to pb4608 50% less often than Owl does
  • Jeff loses to Neuman 100% more often than Owl does.

or

  • Magus loses more often (96% of the time) to pb4608 than to Jeff06 and Owl (84% of the time).

Pasted image740Ă—441 24.3 KB

It is interesting to see how some AIs are better tuned against others :slight_smile:

12 Likes
#2

My AI can beat Owl at 15.5% ? I wonder how it is possible. It’s really good to have this statistics because i was sure my AI was far better than Psycho’s one. But no, we are even.

#3

Surprising eh ? I also really like how evenly matched Jeff06 and Owl are !

#4

Thanks for the statistics. I had a chat with Jeff were we agreed that it was a really cool outcome being that close :slightly_smiling: And you didnt even mention that I beat him in 1v1 by remarkable 0.9 % !!

As a side note: When I woke up 6:00 on sonday, I read an email from 4:00 which said the contest was finished. When I clicked the link in it, the results page listed me as the winner. Really cool … until I saw that Jeffs calculation was still at 60%. 2 hours later, I ended up 2nd. It was kind of sad :frowning:

1 Like
#5

Well I did :> I also really like how evenly matched Jeff06 and Owl are !
I’m really curious to know a bit more about what you did, will you post explanations on the forum ?

#6

I was asked to write something that will get a section in a codingame blog post, I guess.
But it’s really not that fancy. I always analyse those games from a perspective where I look at the overall board (or say state, standing, position…) and how good the situation is for a player in the form of a score. Like chess, where a pawn-lead gives you a base score of 1 while a queen gives 9 and then you can add other factors to it.
The rest is search tree, fast simulation, comparison. And you get your result.

The scoring function is the key.

2 Likes