A disciplined approach to neural network hyper-parameters: Part 1 – learning rate, batch size, momentum, and weight decay:
https://arxiv.org/abs/1803.09820
Like a dictionary?
It’s called PoS (Part-of-Speech) tagging.
No. like:
“I love you”
[pronoun, verb, pronoun]
Hi Jeremy and Rachael any important things we need to know from tensor-flow Dev summit you guys were part of ?
EXACTLY WHAT I’m LOOKING FOR THANK YOU.
Lots of good posts on Medium other than watching the feed on Twitter.
to any mac os x users, you don’t need a 3rd party VNC client.
The built-in Screen Sharing works great
Also, tmux
is a really good, super simple alternative for ensuring you don’t lose your internet connection. sudo apt-get install tmux
. You just call tmux
from the terminal, and it creates another terminal window that is separate from the main one. Do whatever you like. And at any point in time, you can call detach
, and now if you lose your internet connection it won’t matter. Here’s a cheat sheet for getting started with Tmux: http://www.dayid.org/comp/tm.html.
And yeah, screen
is basically the same thing.
Are you talking about “screen” command on the console? Because that stuff is awesome!
tmux wont help when you get disconnected via jupyter.
same for mosh.
if jupyter loses connection, it stops displaying output
tmux just gives you hella windows on Linux and semi-prettier.
What’s everybody running on these days? Is it still predominantly AWS?
Any hope of getting more AWS credits to happen?
Yeah, but your jupyter connection likely won’t go down. That’s running on your cloud’s actual machine. Your personal internet is much more likely to be the break point.
if you are on a weak connection and trying to work on jupyter, and you lose connection, future outputs wont show up (after a certain timeout value).
or if you close your jupyter tab, all your new output is gone forever.
jeremy’s method remedies that.
thanks !
Wow, I though the last lesson had a lot. This one will take me 60 hours to go thru.
I think this was the awesomest class so far(specially as an NLP enthusiast). Thanks so much
Sorry, I’ve probably missed something, but I didn’t quite understood the crux of the lesson. What parts are different from the Part 1 aside from using the pretrained model and using a couple of tricks like approximate sorting?