#metabrainz

zas, samj1912: After additional investigation, I believe that the $inmulti() function is working properly and PICARD-1263 should be closed. I've documented my findings and comments on the Jira ticket. Sorry for wasting your time on this one. You might consider some clarification in the documentation though.

PICARD-1263: $inmulti Script Function Not Working https://tickets.metabrainz.org/browse/PICARD-1263

Leo_Verto: That’s why online learning is a must-have, it’s an adversarial problem, not just a classification problem.

(did not watch the talk though)

they've got an existing overwatch system for trusted players and have plugged their deep learning in there in pretty much the same way that we're planning to do with spam ninjas

the system generates reports and gathers feedback

what I'm a bit concerned about is what features we should train our system on, I'll have to look into more papers on feature selection for relational databases

Leo_Verto: It is easier in MB: spammers only spam, they don’t add any value.

rdswift: thanks for investigating this issue, feel free to suggest documentation improvements regarding scripts

I'll give it some thought and see if I can come up with something meaningful.

yvanzo: yeah but instead of mouse movement and shot data we're dealing with a bunch of complex edit data

Thing is, the documentation makes sense once you understand what's going on.

LordSputnik: Sorry couldn't get to you yesterday, was caught up.

Here's what the summary is https://docs.google.com/document/d/115yIJnO4q4_...

bukwurm: wow, great job writing all that up

I'm kinda leaning towards the second approach - it seems the simplest, and the only real drawback is possible performance issues