« first day (2843 days earlier)      last day (2383 days later) » 
00:00 - 23:0023:00 - 00:00

enumaris
enumaris
23:15
lol
SVC's founder is gently crying in the background
danielunderwood
danielunderwood
It also said something like support vector methods being better for some problems. But they seem to be right for most of the problems I've used tpot on hah
Do you use existing architectures and just modify parameters for your work or do you try to come up with new ones?
enumaris
enumaris
well as far as classifications go...there aren't too many other popular methods...
All the boosting algorithms are generally done on trees
Neural nets + support vectors
what else is there...
Naive Bayes
danielunderwood
danielunderwood
mmm kNN
enumaris
enumaris
depends on the project
sometimes I come up with my own architectures :D
danielunderwood
danielunderwood
N-sided die roll!
enumaris
enumaris
23:20
ah yeah I guess there's nearest neighbor methods...
danielunderwood
danielunderwood
And that sounds fun
enumaris
enumaris
I built a deep residual variational autoencoder
:D
not working as well as I'd hoped so far though lol
danielunderwood
danielunderwood
I want to try my own architectures at some point, but I suppose I should actually figure out how the standard architectures work first
Are autoencoder networks used a lot?
enumaris
enumaris
uhm...not sure about their popularity
I think right now GANs are more popular
(they don't quite do the same thing, but generally speaking it seems GANs has more interest)
danielunderwood
danielunderwood
Do Ng's courses go into them?
enumaris
enumaris
23:24
GAN's? Not that I recall.
At least not in detail.
But his sequence models has Attention-based models which is pretty cool
00:00 - 23:0023:00 - 00:00

« first day (2843 days earlier)      last day (2383 days later) »