Zealous modeler. Annoying statistician. Reluctant geometer. Support my writing at http://patreon.com/betanalpha. He/him.
Friendly reminder that everyone is welcome to join the probabilistic modeling Discord server to discuss all things probabilistic modeling, www.patreon.com/betanalpha/p....
Generative Modeling Discord Server | Michael Betancourt[Despite my best efforts the invite link can sometimes expire. If you find that the link is no longer working then try clearing your cookiewww.patreon.comThe IgNobel goes to everyone who used a black box frequentist estimator, assuming that the asymptotic calibration applied universally.
Randall MunroeCalibration Nobel xkcd.com/3275/
In my recent courses, people have been less and less likely to ask questions, and they usually admit it’s because they don’t want to bother me. The course instructor. In a course designed to maximize opportunities to ask questions. It feels sincere, but it’s definitely a bit of a phase transition.
The Emotion EnginePeople ask AI because it is, to a fault, agreeable and enthusiastic to “help” you and that comforts the universal fear of being seen as annoying BUT what we often forget is any expert who could actually help you is going to be fucking psyched to get into the weeds with you about their field.
Anyone know of any papers that directly critique Markov chain Monte Carlo methods for being unreliable (and not just slow/slower than a competing method)? Papers in stats or applied fields welcome, and the critiques don't have to be well-formed or even correct.
Adrien CorenflosOh I'm actually looking for/collating examples of people (papers) actively shunning MCMC for being unreliable or non-reproducible. Would you happen to know of specific references where this was argued?
People blaming Markov chain Monte Carlo for the consequences of their bad models.
fesshole 🧻I've started saying "fuck sample" instead of "for example" and it improves my boring office job by a fraction.
Choosing to read “through rhetoric, bribery, or blackmail” as how one rides a shark instead of how one convinces a shark. Much more poetic that way.
Josh GondelmanToday I learned that some of my nested quantile visualizations implemented for R and Python in github.com/betanalpha/m... have been implemented for Julia in the FlexiChains library, pysm.dev/FlexiChains..... Neat!
GitHub - betanalpha/mcmc_visualization_tools: Markov Chain Monte Carlo Visualization ToolsMarkov Chain Monte Carlo Visualization Tools. Contribute to betanalpha/mcmc_visualization_tools development by creating an account on GitHub.github.comNothing lasts forever; especially customers. In my latest case study I use survival modeling techniques, including proportional stimuli/hazards by the end, to model customer churn. HTML: betanalpha.github.io/assets/chapt... PDF: betanalpha.github.io/assets/chapt...
Churn, Bayes-by, Churnbetanalpha.github.io<sigh> This is a completely valid way of communicating a baseline and local absolute deviations, which also happens to be the most relevant behavior when most people assess weather across a short time interval. Including zero is important only when _proportional_ variations are most relevant.
turgut keskintürkdata visualization is my passion.
