# Notes

* Slack King und Loses
* dypm keine chains, keine items
* viele Bomben und Wins
* add relation of steps to behaviours
* number of steps per agent in all games aggregated
* divide all behaviours by the aggregated steps
* then compare each behaviour to an arbitrary number of steps (i.e. 100)

## evaluation

### first idea
* compare best agent (A) with worst agent (B)
* try to find behaviours that have a positive impact on A and a negative impact on (B)

### second idea
* compare agents scored first and second
* then compare their found behaviours (with positive impact) with agents performing worse
* good --> worse : worse --> good;
*

### combined
* compare agents scored first and second, try to find behaviour that have a positive impact on both (and negative?).
* then compare their found behaviours (with positive impact) with agents performing worse
* good --> worse : worse --> good;

### general
* of course wins, loss, ties
* compare item_fan, item_thief, item_friend
* compare kick_frency, kick_friend, kick_thief, self_kicker
* compare bomb_layed, hara_kiri, friendly_fire
* compare chain_friend, chain_killer, chain_victim (, chain_fan)