AI libraries


#1

Are the libraries tensorflow, theano and keras on Datalore’s roadmap?


#2

Yes, support for such libraries is on the roadmap. The highest priority targets are obviously tensorflow and keras. Theano has lower priority due to the fact that it’s transitioned to maintenance mode.


#3

Do you plan CNTK support?


#4

Could you tell more information why you want to use it? Why is it better than TF and PyTorch? I.e. what’s the use case where it’s superior to other libraries.


#5

This is the question of portability (or reusing of parts) of existing projects. From my experience, CNTK appeared to be easier and more intuitive for programming. As well, usually I develop a program on a local desktop (or e.g. it can be datalore for collaborative work), after, if necessary, code migrates to the more powerful calculator. In my tasks (scientific use), CNTK appears to be more performant both on single-CPU and multi-CPU (>100-200 cpu) systems.