My personal ML(mainly CV) experiments with various techniques and frameworks.
Advice:
-
Use Python 3 wherever possible. Personal language preference: JavaScript > Lua > Python. Python is good for numerical, while JS and Lua are good for programming in general.
-
Keras rules, but one should always learn TensorFlow first before enjoying Keras's convenience (you always have to hack a few things).
- update: I no longer use Keras, and switched to my own framework, Canton.
-
When in doubt, Use Google. If Google doesn't work very well, use English. If English is not your thing, please give up. You shouldn't be here.
-
Use TensorFlow (Keras' default backend) if you don't want to break things. I run TF on Windows with GPU support.
- update: PyTorch is out, but I'm not switching
-
Keep reading papers. It's good for you!