"Personal use" install commands / notes for attending ML / AI courses.
I am not taking any responsibility! Do not expect me constantly tracking the compability issues!
cmd.md: Command lines / notes.mujocopy.md: deprecated Topics on installing and runningmujoco-pyon Windows / WSDL. Current version of mujoco (as in 2201) has been fixed the issue.fcnd.md: Special for FCND environment.cmd.pyapplies here.vscode.md: Topics on VSCode integration.cmd.pyapplies here.
- Not joking. You must have all hardwares / softwares in recent generation and not fall too far behind the mainstream. Otherwise, you won't even able to get the install packages!
-
Intel 4th gen or later. Most libraries needs FP16 / FP32 instructions, which is under Intel x86 CPU with AVX2 instructions, i5 2500K would be drastically slow compared with gaming. AMD/ ARM is not tested. Should be better / newer than AMD Bulldozer *cough and Nvidia Tegra *cough?
-
Nvidia GPU with Pascal or newer. GTX 10 series for general publics. This article will use TF 2.5.0 under this official tested build configurations. Also, for CUDA inside WSDL2, it is a hard requirement also. Maxwell (as extreme as 750 Ti, but not 780 Ti or Titan Z) is "marginal". See here and here for corrosponding CUDA / GeForce driver requirement. For other GPUs (AMD, Apple M1), I never tested this approch for AMD or this approch for Apple M1.
-
Sturdy RAM, SSD and HDD. Your dataset will rushing in & out in GBs. "OEM / Stripped server parts" is nice, but "rebuilt SSD / RAM" is too risky. Location / brand is not important, but do get from reliable sources (for my case, extensive testing under 7-day warrenty from 2nd hand sellers).
-
Powerpul cooling. It is worse than 24/7 mining! Keep monitoring the temperatures with software. If you can't keep it in pace, consider Disable Intel Turbo Boost, and be patient.
-
Windows 10 1909 or newer. You may need WSDL2 for Linux subsystem.
-
Linux Ubuntu 18.04 or newer. as part of WSDL2.