Monaco: Quantify Uncertainty & Sensitivities in Computational Models w/ Monte Carlo Lib | SciPy 2022

Roll the dice! Quantify uncertainty and sensitivities in your existing computational models with the “monaco” Monte Carlo library. Users define input variables randomly drawn from any of SciPy's statistical distributions, run their model in parallel anywhere from 1 to millions of times, and postprocess the outputs to obtain meaningful, statistically significant conclusions. This talk will go over why you should always be running Monte Carlos, a demo of how to set up and run a sim, and a crash course in generating relevant plots and statistics.
Project repo: github.com/scottshambaugh/monaco
Lots of examples: github.com/scottshambaugh/mon...
API Documentation: monaco.readthedocs.io
Conference paper: conference.scipy.org/proceedi...
Talk slides and notebooks: github.com/scottshambaugh/mon...

[@wsshambaugh]

Speaker here! I need to work on my "ums", but I'm pretty happy with the content of the talk. Hope it's useful to some people! I can field more questions in the comments here.
The Q&A questions were poking at monaco's place in the ecosystem of similar tools. In general, I wanted to have the focus of monaco be making forward uncertainty propagation as easy and straightforward as possible, with sane and rigorous defaults. And of course given the venue, it and the other tools I mentioned had to be open-source python.
I don't have much experience with the tools asked about in the Q&A besides a quick look at their docs and examples (the ones I've used/written before have been internal to companies), but here are my hot take initial impressions:
* Dakota - Like many of Sandia's tools, very powerful but clunky to use. Also not python!
* OpenTURNS - Looks great, much broader scope than monaco's. I would put it in the same class as UQPy.
* UQLab - Looks great, but not python! If I was in a matlab company, this would be my pick.
* BATMAN - Similar to monaco's focus, but no longer maintained. If I had found this before starting my project I might have just contributed to it instead. The author has done great work on the scipy quasi-monte carlo module and elsewhere!

[@alexlanderos5299]

Hello, I listened to your talk during scipy and have revisited it often so I wanted to say thank you for the great talk and inspiration! Im an undergrad aerospace student working on a liquid biprop rocket project and I recently modeled and created a UQ wrapper for our rocket in a matlab/simulink setting and had I seen your comment I might have used UQLab instead lol. Best of luck in your line of work (which is awesome by the way) and future development!

[@wsshambaugh]

@@alexlanderos5299 Glad you found the talk useful! Good luck on the biprop, they can be finicky to get working right!

[@theMou]

Thank you!