Hey /g/,
Currently fiddling with some machine learning software. Got it running on the raspberry pi that I use for hobbyist projects, but, as expected, the training is too slow to be practical. I need an inexpensive PC with lots of RAM (4gb is the absolute minimum) and a decent GPU. The cheaper, the better — I'd like to buy multiple machines to test different datasets.
Any recommendations?
Happy to answer questions about my project and/or have a general discussion about machine learning as well.
The GPU is going to be the hard part. The computers themselves could probably be gotten on Craigslist from businesses purging old workstations. That's a reliable source of machines for pretty much any practical use.
>>61491602
Surplus is an interesting idea, thanks for sharing. After doing a bit more reading, I've found out that Android actually supports Torch (the machine learning library I'm using), so used phones are also an option. There's a precedent for this kind of thing, since droids have been used for research in the past.
What's the dev environment like on Android?
>>61491573
I would have recommended the fury if your dataset will fit in 4GBs of RAM, it has some freakish compute for 250 bucks...
Assuming the mining crisis hadn't emerged.
>>61491573
what ML r u doin fämäläm
Agree with buying an old workstation. Dual xeons on old sockets can be had cheap. Get an older Nvidia GPU if you can.
Im just interested, what will your code learn?
>>61492404
Text-based stuff. End goal is to have a bot that can write & recite hip-hop. That plus a library of beats should make for a pleasant 24-7 music service.
Working mostly off the code here:
https://github.com/karpathy/char-rnn
Wrote my own scripts to mine websites for lyrics, and I'm currently playing with ways to make the text-to-speech rap on beat.
If this ever gets fully realized, I'll come back and share the results.
>>61492322
What exactly is "the fury"? I'm behind on hardware, sorry.