Feeling absolutely delighted at this snippet from my work-in-progress real-time physics #simulation -with- #visualization project.

That's TWO HUNDRED AND FORTY separate balls bouncing off each other 500 times per second. It's fun to have an excuse to optimize #Python (#Numba and @matplotlib ).

code for the curious:
https://github.com/brohrer/how-to-train-your-robot/tree/main/chapter_5/sandbox/al_falling

Feeling absolutely delighted at this snippet from my work-in-progress real-time physics #simulation -with- #visualization project.

That's TWO HUNDRED AND FORTY separate balls bouncing off each other 500 times per second. It's fun to have an excuse to optimize #Python (#Numba and @matplotlib ).

code for the curious:
https://github.com/brohrer/how-to-train-your-robot/tree/main/chapter_5/sandbox/al_falling

Did you know that #sklearn pipelines that rely on k-nearest neighbors graph computation can swap the built-in exact neighbors search by an accelerated third-party method such as the one implemented with #numba in the #PyNNDescent project?

https://pynndescent.readthedocs.io/en/latest/pynndescent_in_pipelines.html

#PyData #MachineLearning #Python

@aldanial Our signal processing code is already doing parfor in #MATLAB, but its good to know we don't have to give up the parallellization by calling a "#python MEX function". We are definitely CPU-bound and IMO MATLAB-bound and this could alleviate both problems.

I've never used #numba, I'm more of a #numpy person. Are they equivalent?

Excellent presentation by Max Mergenthaler from Nixtla at #PyDataNYC on bridging classical and deep learning methods for #timeseries forecasting in #python thanks to #numba

Code: https://github.com/nixtla