Closed andrew-son closed 4 years ago
All examples of this in the dataset are suicides (i.e. attacker and victim are the same player), which can happen when you type "kill" in the console or when players crash/disconnect when they are alive. We already filter rounds which were restarted, so either this happened and had direct impact on the match, or our heuristic for filtering missed an edge case.
I'll go through the cases and see if I can figure it out. I'll probably also filter kills in freezetime, as they are usually match medic type of stuff.
@kristoff3r If it somehow helps: basically what I did was filtering all rounds with kills more than 10 and I found out entries with like 10 kills at spawn points so I thought that's trailing suicide kills from round restarts (since they were marked as team kills)
As far I understand, "World" kills are mostly round restart kills (like in entries with 10 team-kills in
previous_kills
withattacker_side
andattacker_position
fields present; positions seems to be spawns) or self-kills (in that caseattacker_side
andattacker_position
fields areNone
). However, one snapshot differs from both cases (index 68121 in combined dataset). It has bothattacker_side
andattacker_position
fields set and the position is not a spawn position (tunnel stairs to be exact). Is it a bug? Can we just safely omit all the "World" kills except ones withNone
attacker_side
andattacker_position
fields, which seems to indicate a self-kill (e.g. from fall damage, not sure iftrigger_hurt
on maps like vertigo counts)