Luis Damiano

Bayesian Stats, Probabilistic Machine Learning, R & Stan, Open Source, and any linear combination of whatever may help me push the envelope in this needy world.

[Luis Damiano @ldamiano0]

Next time somebody asks "what do you for a living?"... This week I attended the @TheSIAMNews Uncertainty Quantification 22, here are all the +400 abstracts summarized in a word cloud. Code: https://t.co/643nH7FUC1

#SIAMUQ22 #Statistics #DataScience #R #RStats #AcademicTwitter

[Luis Damiano @ldamiano0]

@eddelbuettel @joshua_ulrich after the improvements that base R has seen over the years, in which situations it's still worth using C++ instead of R directly for speed?

[Luis Damiano @ldamiano0]

@joshua_ulrich Run the same matrix operations in two computers, local & remote. Couldn't use microbenchmark, so went with system.time instead. Which of user.self, sys.self, and elapsed is conceptually the closest to the time that microbenchmark measures? #rstat

[Luis Damiano @ldamiano0]

@jasonschreier PC, loved Stardew Valley and generally enjoy simulators unless they become too grindy like Graveyard Keeper.

[Luis Damiano @ldamiano0]

@SalomoneRob I came across both uses of VI in the context of GPLVMs and Deep GPs. I'm learning VI for computer experiment emulation, so "standard" VI is what I'm most interested about (at least for now).

[Luis Damiano @ldamiano0]

@SalomoneRob what read would you recommend for learning Variational Inference? I'm planning on use it on Gaussian Processes but I'm also interested in general canón material as well.

[Luis Damiano @ldamiano0]

@SalomoneRob Thanks, this is great! Definitely gonna check out these resources. Like R for everyday tasks but can do Python, better be able to handle both these days :)

[Luis Damiano @ldamiano0]

Midwest fall is quite something... been missing this

[Luis Damiano @ldamiano0]

@uPicchini @matvil My goto reference book. It's all probabilistic, can't help trying it first and then Bishop if need be only. ML is evolving rapidly so a recent book has a clear advantage, but even 2012 smells old. I see them as a warmup/refresher/bkgrd before getting into the papers of interest

[Luis Damiano @ldamiano0]

@smartin2018 I need multiple code views so much in RStudio :( Also some better refactoring tools, I'm starting to feel that it falls short once to get into developing a large codebase.

[Luis Damiano @ldamiano0]

@MaartenvSmeden 1. none if it's a p-value, rather avoid that; 2. depends on model accuracy and the faith you have that it holds (usually close to none); 3. depends on domain-specific knowledge, sometimes 0.5 and 0.6 are the same; 4. also try to keep consistency with other figures & tables.

[Luis Damiano @ldamiano0]

@gztstatistics Just ask the man, @joshua_ulrich is the main maintainer of microbenchmark among many other packages.

[Luis Damiano @ldamiano0]

@gztstatistics Maybe align the y labels so that the three equal signs in each line all match.

[Luis Damiano @ldamiano0]

@betanalpha Awesome, makes more sense now. I think so far I've been thinking of model identification in terms of the joint pdf and never considered it from the point of view of the marginals, which are a byproduct of the joint. Thanks for the clarification, great insight!

[Luis Damiano @ldamiano0]

@betanalpha Or are we talking about two different model specification issues?

[Luis Damiano @ldamiano0]

@betanalpha I never considered (or realized) that our conditionals may not be enough to identify the **marginals**. So it's not about the joint not existing, or having a shape too hard to explore, but about the marginals not being in uniquely determined?

[Luis Damiano @ldamiano0]

@NeuroStats @f2harrell @MaartenvSmeden @venkmurthy @Research_Tim @moritzkoerber @ESteyerberg @ReneEijkemans this discussion on having methodologically sound external validation caught my attention, do any of you have any read to suggest? I'm looking for something more in depth/insightful than a typical low level textbook overview of bootstrap and k fold validation

[Luis Damiano @ldamiano0]

@NeuroStats @f2harrell @MaartenvSmeden @venkmurthy @Research_Tim @moritzkoerber @ESteyerberg @ReneEijkemans I'm not surprised, I haven't found a good read on this. Most texts will just define, give a one-paragraph indication on how to use, and give two examples about these methods. But as far as the lit I've read, none discuss in detail & in an unified way when to do what, how and why.

[Luis Damiano @ldamiano0]

@gztstatistics I'm interested in this, did you find anything worth mentioning?

[Luis Damiano @ldamiano0]

@dan_p_simpson When are you coming to ISU, pretty please?

[Luis Damiano @ldamiano0]

@vianey_lb Sounds like a great rule, I'll keep this in mind. Actually, it reminded of exposure therapy (https://t.co/9svriDRSSC), exposure therapy for academics.