sup /g/irlfriends I've been doing machine learning shit

>>61757004
Don't do it on the laptop. You can try some stuff first on AWS GPU spot instances first, and if you really want to get into it and STICK to it, then build your personal workstation.

>>61757004
Also Jupyter Notebook is great for remote machine learning purposes.

>>61757029
oh yeah I forgot about jupyter

how does that work? as far as libaries, so far I've just done some skiddy shit with tensorflow and everything else by hand with numpy

can you basically setup your desktop as a server and connect in through jupyter?

>>61757074
>can you basically setup your desktop as a server and connect in through jupyter?
Yes, and it's very easy. If the server and your laptop are in the same LAN, just do on the server jupyter notebook --ip=0.0.0.0, and then connect on your laptop via a browser to 192.168.x.x:8888

>>61757115
should be gucci if I'm at a coffeeshop or workspace and port forward beforehand though, right?

>>61757004
You can do a lot on a shitty 5 year old laptop.

I got a 1070 when I started toying with DNNs and it serves me well, the 1070 is the sweetspot hobbie ML card ATM.

>>61757152
Personally I'm using SoftEther VPN in that case (connect to the server via the VPN and you're in the same LAN), but other solutions could work. There's also the choice of opening the notebook to the public internet with a password authentication, but for safety reasons I wouldn't recommend it.

>>61757174
my desktop has a 1080, I think that'll be good right?

>>61757215
Then again that depends on what you are doing, you could make a trivial system that would take centuries to finish in even the world's strongest supercomputers.

I made a "perfect" pong player with my 1070, it took 19 hours with a very greedy reward.

Good ML cards should have a lot of RAM because when they crunch big matrices they need to have them loaded in memory.

In general, most operations in ML at this level are memory-bound, so the performance of a 1070 and a 1080 are about the same, the same goes for the 1080 Ti and Titan cards.

Also if you are going AMD, make sure beforehand that the library you want to use has OpenCL support, most don't.