visual-studio-2015cntk

Learn how CNTK 2.0 Beta 8.0 source code works on Windows 10 using Visual Studio 2015?


Problem

I am a student, .NET developer, and I would appreciate the advice of heavy weight contributors and designers of CNTK. I am going through the documentation and the tutorials and the actual source code is impressive.

What is the best api to use in visual studio 2015 so that I can step into the functions in CNTK to see how a multilayer perceptron has been coded up by Microsoft? Often times, with .NET projects I write a sample application I understand and then just go through the source code while on the debugger. This is the best way for me to learn. My question is for visual studio 2015, but I am willing to step out of my comfort zone and learn if best practices is to do this on a different IDE and even OS.

I plan on using the tool as is, but I would like to become as familiar with the tool as possible because I would like to start using it over Caffe, TensorFlow, and etc since I am a .NET developer.

Attempts

I have read through tutorials on CTNK and I find myself using the brain script a lot, but I often invoke it on the commandline. I know there is a python api in visual studio 2015 and I was wondering if this is the preferred method to getting to the backend with CNTK?

Please do not down vote this. I've read the documentation that is provided and I did not feel it was appropriate to ask this on GitHub repository and this is the only other venue I felt appropriate to ask this question. If there is another channel or forum to ask CNTK general questions, please let me know and I'll do so, but if not I appreciate your mentorship and kind assistance to a earnest learner.


Solution

  • This place is a good place to ask questions about CNTK. I'm sure you find lots of support and advice here.

    VisualStudio is a very good environment to explore the code base, it is the dev environment most contributors use

    regarding the best way to control cntk you will find many different opinions. Brainscript gives you excellent / script like control, although many people will point to Python as the preferred way to communicate with CNTK. There is certainly a lot of general Python support in the ML community!

    We just started exposing C# on the eval side in the latest betas, and you can expect more support for .net languages is coming, but also other languages might show up ...

    I don't think you can pick one language as superior to others, usage is often similar, and in the end it depends on your personal preferences, experiences, and certainly the problem you are trying to solve

    Wolfgang