jboelaert
jboelaert
Thanks @paulgirard this is great. Here are my takes # Top selects Very good points, this is an heirloom from the clunky old R version - The real logical problem...
Exactly!
The Lasso algorithm is different, more efficient in high-dim cases. And if we advertise Lasso in the UI, it's better to call an actual lasso
Ok, actually in sklearn Lasso is only for (numeric) regression, not classification... So, new suggestion: - replace Lasso name by Linear-L1, and in backend LogisticRegression, l1 and lbfgs - replace...
Indeed, saga seems to be the only viable option for multiclass with L1. While we're at it, I see that sklearn.LogisticRegression also takes a "class_weight" argument to handle unbalanced classes,...
Good idea, but probably hard to implement: - can't really be in main tagging window, because it requires having the Bert model in memory - a BERT predictions exploration tab?
Indeed, I can't reproduce it either today on the PC. It doesn't work on my android firefox though
Also, I just got what took me time to understand on PC: click-and-drag doesn't do selection, while it is the universal way in OS. In the current states it moves...
Quick update: from France with an optic fiber connection I still have the same problem, about 12 sec for 80+20 obs.