if you have access to matlab it was by far the easiest to understand and the tutorials are top notch. Matlab is stepping up its game in the neural network world. Its also relatively easy to deploy as you can compile a matlab trained net into an exe which will take a certain amount of work in say tensorflow or any other thing with minimal documentation.
The allure of tensorflow is that is sooo powerful , google uses it, but the trick is that the amount of tutorials/quality up-to-date guides on how to make anything remotely useful and deploy-able are intentionally scarce. If you are a novice, I would stick with matlab, R, and good old C, things like code-projects have amazingly good tutorials on how to write your own code for learning. The thing is once you write your own library/basic framework you have a tool that is all yours and you understand how it works. I would abstain from blindly git pulling and running code as the amount of time you will spend trying to figure out where it broke is an order of magnitude more than writing your own simple code when it comes time to applying it to your field/project. Just what I learned through time