Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

Recently I had a convo starting with: "Maybe we can use AI to infer whether we are dealing with an experiment or a control based on the metadata of these public studies."

These data are tables, people call "controls" anything from "control" to "ctrl" to "ctr" to "t0", either in the file name, a random column, etc etc etc. It worked well and I'm glad we tried it. In time I think we will derive value from deciding to use it. I'm glad nobody tuned out.



I mean I am not really talking in a professional setting or technical setting... I mean just shootin the shit with people.


I mean, it sounds like you could’ve had a few if/else statements and accomplished the same thing.


Yeah using vectorization for this seems like bringing a cannon to a duel. At most I would've used levenshtein distance or something




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: