The Doctor
Senior Member
- Dec 18, 2010
- 1,110
- 474
I'm doing pretty good with Tensorflow. I can think of some pretty sweet IM ideas for ML so I'm heading towards that. The maths are annoying but not impossible.
There is a person on Youtube who goes by the username Sentdex, he makes a lot of videos on python and actually has a series about machine learning. youtube(dot com)/watch?v=OGxgnH8y2NM&list=PLQVvvaa0QuDfKTOs3Keq_kaG2P55YRn5v. I found it to be really awesome. Hope it helps you out.
oseart
Actually I can't PM you. Profile isn't old enough. PM me?
Also about training time: You will want a good NVIDIA GPU. I bought a TItan X Pascal just for ML. You don't NEED a TItan X but you should have at least a GTX 1080. 1060 or 1070 will work but will take more time to train. I'm adding a second GPU soon. They don't have to be the same model if you want to have more than one and SLI is not required. They do have to be NVIDIA to use CUDA. That or you could use Amazon GPU compute clusters and just pay a few dollars each time you want to train (Don't forget to destroy your instances/cluster when you're done!).
More on the maths: You can get by without the maths for a while just using the libraries BUT until you learn the underlying algorithms (Which has maths involved) you won't be able to tweak the algorithms or even make your own variations of them. This can sometimes mean the difference between 95% accuracy and 98% accuracy. That may not seem like a lot but of course it is.
Also, I suggest learning: Python, Nim, MATLAB, and R.
On the plus side: If you learn these things you will be part of the elite group of people with the security clearance. These skills, IMO, are essential for the coming years. It's my opinion that people who don't learn these things will soon find themselves without a job. Maybe they'll get paid some standard income but the ML coders will have a spacial place in society.