This isn't really anything against W&B- it's very nice and polished. I'm really just advocating critical thinking about your process.https://twitter.com/citnaj/status/1320443311845302272 …
-
-
Replying to @citnaj
I think if one is going to use an experiment tracking UI, W&B might be the best choice. But *all* experiment tracking UIs I've seen are really distracting, and aren't a replacement for thinking carefully about what you actually want to analyze, and the best way to analyze that
3 replies 3 retweets 28 likes -
Replying to @jeremyphoward @citnaj
Pretty much every early practitioner I've watched, however, uses an experiment tracking UI. They're very tempting!
4 replies 0 retweets 4 likes -
Replying to @jeremyphoward @citnaj
What do you cats think about hyperparam tuning like optuna and raytune? Esp in terms of real world problem solving vs kaggle comps.
2 replies 0 retweets 1 like -
Replying to @waydegilliam @jeremyphoward
Just speaking for myself, I'm inclined to just do enough tuning on hyperparameters so that I'm not shooting myself in the foot with something super sub-optimal. But the bigger and more worthwhile gains have been in focusing on the bigger picture stuff, imho. 1/
1 reply 2 retweets 31 likes -
Yeah, I'd say part of what makes hyperparameter tuning so seductive is that it keeps you busy enough to avoid having to think about the big picture, while still being able to justify your time spent on this. Just like in that classic
@xkcd on compiling
pic.twitter.com/m74HaEoDwi
3 replies 7 retweets 61 likes
Wow, yes, I think this is a great way to put it!
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.