Battlegrounds Compositions Power stat - Zero-to-Heroes/firestone GitHub Wiki

I'm playing around with various ways to measure a comp strength from the data.

A difficulty compared to building stats for decks or heroes is that the comp is not fixed throughout the game, so you first need to know at what point to measure the data. I'm for now simply taking the last known board state.

But the main difficulty is that some compositions come online later than others, because some comp rely on tier 5-6 units while others can rely on tier 4 (like the naga spell comp). This means that you could have part of a Naga Spell comp by turn 6. If you lose, you have a turn 6 loss attributed to Naga Spell Comp. But if you plan to go Buff Shop Demons that require tier 5 units, and you lose turn 6, your comp will not be assigned to a Buff Shop Demon (because you don't have the pieces we need to identify the comp yet).

This means that some comp will naturaly have better average final positions, because all the early defeats won't be counted towards that comp.

What I'm using instead for the "Power" rating is:

• counting only games where you make it to top 4
• what's the share of the games where a given comp finishes first?

The idea behind it is that, when you are top 4, your comp is much more likely to be set (though that's not necessarily true). And so a loss is more likely to be attributed to the correct comp. At the same time, top 4 is not so restrictive that very greedy comps have a strong positive bias.

What do you think? I'm curious about your thoughts on this