use np.cumsum and eliminate inefficient loops

[maldil]

Hi Maintainers of this repo,

Thank you very much for your excellent work.

I am new to this repository. I am a researcher studying the best practices of evolving data science codes. According to our findings after examining 1000 data science repositories, migration of loop-based calculations is a widespread evolution practice among developers since it improves performance and code quality. I noticed in this PR's code that it might make better use of NumPy functions and avoid unnecessary loops.

This PR is a minor contribution compared to all the hard work that you have done in this repo. However, I am hoping that it will enhance code quality and, hopefully, performance.