Torch.jl VS oorb

For frontends Python has quite some head-start. In principle it would be possible to write Julia frond-ends to existing ML libraries (written e.g. in C), for example https://github.com/FluxML/Torch.jl , but the advantages over Python frontends would be very limited. Only a front-to-back Julia implementation leverages most of the language advantages like composibility and flexibility.

PyTorch for example is a C++ library with a Python user interface, see e.g. the language shares in GitHub (https://github.com/pytorch/pytorch ). There is also a Julia binding for Torch (https://github.com/FluxML/Torch.jl), but I do not know how up-to-date it is.

Imagine my surprise to see OpenOrb, the standard open source software package for orbit determination and minor planet propagation, on the front page of HackerNews. Its interesting software with a beautiful theoretical basis in Bayesian statistics, and a gnarly Fortran codebase - I can’t wait to see the discussion!

Oh.

It’s one thing to land near a name in use. It is quite another to take it directly!

https://github.com/oorb/oorb

Mostly Python and Fortran. See for example https://github.com/oorb/oorb.

The hardest problems are always the social ones. How do you get uptake of a new method, how do you get funding, how do you politely tell a collaboration they are doing the wrong thing, etc.

But if you mean pure technical stuff - the hardest problem I had to solve was rethinking some of the inner loops of the THOR algorithm. The problem was essentially to speed up a Hough transform in 6D space. Lots of time spent profiling CPU cache timings to get that fast.