It’s better than the rest at the moment. Let’s say you have small amount of data, 1Go than can easily fit in the memory of a laptop. You try to load or do ana

It’s better than the rest at the moment. Let’s say you have small amount of data, 1Go than can easily fit in the memory of a laptop. You try to load or do analysis of that with python. The only way to make it work is to use a library that is build in C. Forget doing for loop over that thing. So each library has to build a language of parameters to allow you to use the speed and memory efficiency of C. Effectively in ML python act as a config language for C. So all the actual work has to happen in C, but all your experiment are in python. That create a barrier to innovation as you hit limit of performance on python and you have to switch language. But yeah, all the library are in Python and most of the work is still done there so it’s very hard for people to move to something else because they would have to rebuild a huge part of the stack, there is a strong network effect here. But the core language is still very bad for performance computing.