What is the current state of C# with machine learning (problems and complexities)?

Question

C++ / C#

Greetings!

For comparison, let's take the most popular machine learning language Python + ML libraries.

1. What libraries do you use/use under C#?
2. In comparison with Python, which machine learning methods C# does not allow to use (or are not implemented conveniently/difficultly), which, in turn, are very important in machine learning?
3. What is the situation with GPU training?
4. Performance?
5. Your answer is that "not so", and in general I will be glad to any opinion.

Answer 1

1 python is an interpretive language, slow. to build a neural network, it must be compiled. for the learning task, the speed of the language is not important, since the main time is occupied by the neural network. Python's popularity is due to its simplicity.
2 s allows you to do whatever you want.
3 of course, the neural network is better done on gpu, this is the main direction. there are also specialized chips for neural networks, for example, Google released a tensor calculator
in general - machine learning is almost independent of the language. many libraries. choose the one that is more convenient for you.

Answer 2

What libraries do you use/use under C#?

ML NET, Tensorflow net

What is the situation with GPU training?

It seems that Tensorflow net can

Performance?

Seems ok

Your answer is that "not so", and in general I will be glad to any opinion.

Personally, it doesn’t suit me in ml net that it doesn’t allow you to design NS by hand - it decides what the architecture of the neural network will be, according to the input data.
I also don't like that the image can be submitted as input only as a path to the file.