Notices by Eden (edensaesthetic@witches.town), page 6

Also realtalk catboost.yandex is becoming one of my favourite go-to algos.

I usually go straight to xgBoost because it's really simple to do advanced tunes in it, but catBoost is starting to outperform it on the kind of predictive models I usually do - mostly like marketing sales junk.

Which is like HYPE.

But also. I am now doing a deep learning mechanism on it just for fun and no longer have justification for doing it at all. :')

WHICH SUCKS.
because R just released the keras package that makes it super easy to deploy tf models, like literally

model %>%
dense_layer(activation = 'relu')

Is the kind of baseline difficulty?! And it's not particularly quick, but it's dead simple to test things before writing it in something better.