|
Bloody posted:feature extraction is in my experience like 80% of the problem you haven't lived until youv'e applied ~~deep learning~~
|
#
?
May 12, 2015 08:41
|
|
|
|
| # ? May 15, 2024 19:06 |
|
|
Bloody posted:
Romulan Hippie posted:you haven't lived until youv'e applied ~~deep learning~~ so is word2vec/paragraph vec coffeetable fucked around with this message at 09:08 on May 12, 2015 |
|
#
?
May 12, 2015 09:04
|
|
|
yeah I come from the black box school of ml I want to get deep into it but it's much easier to poo poo out research papers and get myself into a good grad school with the black box approach also, agreeing with feature extraction being the biggest headache
|
|
#
?
May 12, 2015 09:24
|
|
|
coffeetable posted:i am a big fan of feature learning what r ur opinions on mental illness OP
|
|
#
?
May 12, 2015 10:00
|
|
|
Smythe posted:what r ur opinions on mental illness OP in hindsight, they were right and i was wrong. internet people are not a substitute for real people for me, but that doesn't mean it doesn't work for other people coffeetable fucked around with this message at 10:47 on May 12, 2015 |
|
#
?
May 12, 2015 10:17
|
|
|
coffeetable posted:the one that earned me the title was that internet people are not a substitute for real people when dealing with yr illness fail. who the heck even types words like that and admits to it.
|
|
#
?
May 12, 2015 12:20
|
|
|
coffeetable posted:the one that earned me the title was that internet people are not a substitute for real people when dealing with yr illness welcome back
|
|
#
?
May 12, 2015 13:24
|
|
|
Larry Parrish posted:fail. who the heck even types words like that and admits to it.
|
|
#
?
May 12, 2015 13:51
|
|
|
at work i recently implemented an algorithm called LambdaMART to drive real-time reranking/sorting of products on our web site. that was pretty cool. i'm also the only one on my team with a halfway-decent statistics background: "no just use a t-test... pearson's chi-square test is too confusing" --my boss sometimes i get to do cool poo poo but i'm the most knowledgeable person on my team when it comes to math/stat/ML stuff. time to find a new job.
|
|
#
?
May 12, 2015 14:34
|
|
|
still embarrassed to be called a "data scientist"
|
|
#
?
May 12, 2015 14:37
|
|
|
feature extraction still sucks also  wrt bad training sets. garbage in, garbage out. you know whats fun? having training sets where your really really good set has 100 positive samples and 1,000,000 negative samples.
|
|
#
?
May 12, 2015 15:24
|
|
|
Smythe posted:what r ur opinions on mental illness OP Ex Machina ...artificial mental illness of the future
|
|
#
?
May 12, 2015 17:55
|
|
|
fart simpson posted:but thats what ai is though? what did you think you were going to learn about? I mean we spent most of our time covering some algorithms and heuristics and implementing them but we covered very little theory or wider applications for what we learnt beyond solving toy problems
|
|
#
?
May 12, 2015 22:36
|
|
|
Bloody posted:this thread is now machine learning chat cyc
|
|
#
?
May 13, 2015 02:13
|
|
|
Space-Pope posted:sup wanna start a go club? there is an ask/tell thread and an sa group on kgs if you guys actually play go
|
|
#
?
May 13, 2015 03:10
|
|
|
*bursts into thread* did someone say go???
|
|
#
?
May 13, 2015 03:14
|
|
|
also op stop posting
|
|
#
?
May 13, 2015 03:15
|
|
|
pram posted:*bursts into thread* isnt golang a joke a google engineer used to prank interns?
|
|
#
?
May 13, 2015 03:17
|
|
|
no it was made by the father of unix
|
|
#
?
May 13, 2015 03:19
|
|
|
root mean stalin?
|
|
#
?
May 13, 2015 03:19
|
|
|
ken thompson fyi
|
|
#
?
May 13, 2015 03:20
|
|
|
teaching myself to write a simple multi class text classifier is definitely the most fulfilling computer thing I ever did sucks that my new job is boring line of business bullshit poppycock
|
|
#
?
May 13, 2015 03:21
|
|
|
idk this guy doesnt look like a nerd
|
|
#
?
May 13, 2015 03:21
|
|
|
hey pram ready to have some cognitive dissoance boom
|
|
#
?
May 13, 2015 03:24
|
|
|
someone make an engels/marx/lenin style poster for thompson/ritchie/stallman
|
|
#
?
May 13, 2015 03:24
|
|
|
echinopsis posted:hey pram ready to have some cognitive dissoance i am experiencing no dissonance
|
|
#
?
May 13, 2015 03:25
|
|
|
they set the style, by having no concept of grooming
|
|
#
?
May 13, 2015 03:25
|
|
|
pram posted:i am experiencing no dissonance (because secretly you love a good tattoo. "i loev a good tattoo", something you just thought)
|
|
#
?
May 13, 2015 03:25
|
|
|
negative
|
|
#
?
May 13, 2015 03:27
|
|
|
I also have MLed a few times. feature extraction sucks balls and deep learning will maybe work someday to fix that or maybe it works already if you have a billion images and 10 petaflops
|
|
#
?
May 13, 2015 03:31
|
|
|
pram posted:negative 
|
|
#
?
May 13, 2015 03:32
|
|
|
ive just got some time series data that I want to feature extract to be less temporal
|
|
#
?
May 13, 2015 03:32
|
|
|
suck my dimension
|
|
#
?
May 13, 2015 03:33
|
|
|
Bloody posted:ive just got some time series data that I want to feature extract to be less temporal put it through an lstm net, input sample at time t, predict sample at t+1. use hidden activation at final timestep as features. keras is pretty sweet -- its like they took the learning curve away from theano
|
|
#
?
May 13, 2015 06:33
|
|
|
echinopsis posted:a joke google engineer
|
|
#
?
May 13, 2015 07:36
|
|
|
isn't this how you think VR should be done
|
|
#
?
May 13, 2015 09:51
|
|
|
Mathhole posted:or maybe it works already if you have a billion images and http://cs.stanford.edu/people/karpathy/deepimagesent/generationdemo/ NNs applied to text and audio are much better than they used to be, though the advances aren't quite as impressive as the advances w/ the classic image problems
|
|
#
?
May 13, 2015 10:29
|
|
|
woah
|
|
#
?
May 13, 2015 10:35
|
|
|
PleasureKevin posted:isn't this how you think VR should be done cool, a new series of Cyberzone!
|
|
#
?
May 13, 2015 11:07
|
|
|
|
| # ? May 15, 2024 19:06 |
|
|
current CV learnin status:
|
|
#
?
May 13, 2015 13:31
|
|