Halite Home

Distribution of sizes and num_players?


Maybe more for strategy rather than rules, but @Sydriax or @truell20 or any experts, what is the correct joint distribution of game sizes and num_players? asking cos pulling the games encountered by my bots gives a different distribution than for example the diamond replay dump from @andrewyang96 . [Ed: that could easily be because of small sample size, so easier to ask for the distribution] thanks !

Concrete info : achieving top 50 via ML (Policy Gradients)

Map Sizes: [20, 25, 25, 30, 30, 30, 35, 35, 35, 35, 40, 40, 40, 45, 45, 50]

Number of Players: [2, 2, 2, 2, 2, 3, 3, 3, 3, 4, 4, 4, 5, 5, 6]


One note, as originally pointed out by @erdman, the above player number distribution only applies to the seeding player and will be different for non-seeding players.

This is a direct result of, a) having some games with more than 2 players, b) choosing one player as the game 'seed' and c) having a non-uniform distribution in the choice of the 'seed'. If you look at only games where the player is the seed it will follow the above distribution. But games where a player is pulled in will be distributed more toward the larger player numbers. c) above doesn't actually change this difference, but it does mean that since you are much more likely to be a seed when the submission is new the distribution will change as your likelihood of seeding goes down (currently there is a sharp change after 400 games).


For local testing, I use this distribution of player-counts to simulate non-seeded games:
[2] * 5 + [3] * 8 + [4] * 9 + [5] * 8 + [6] * 5

The distribution of map sizes does not change w.r.t. seed vs non-seed.


Thanks all for the detail. It was the non-seeded thing which was causing the diff, i'd never have spotted that in 20 sundays! tks vm