Hyper-specific technical stats question that will take a little bit of setup: Consider J parameters Mu = (mu_1, ..., mu_J) which are partitioned into two subsets, Mu1 and Mu2...

Hyper-specific technical stats question that will take a little bit of setup: Consider J parameters Mu = (mu_1, ..., mu_J) which are partitioned into two subsets, Mu1 and Mu2...

“Fear not the those without papers, but those without conscious.” www.instagram.com/p/DT8OwTxjSbm/

mods are asleep post forbidden greek letters

Fitting a machine learning model.

On Tues, February 17th I'll be giving a talk and short workshop on (narratively) generative modeling for Princeton's psychology department, psychology.princeton.edu/news-events/.... The talk (and maybe the workshop) will be live-streamed for those interested in attending remotely.

On Tues, February 17th I'll be giving a talk and short workshop on (narratively) generative modeling for Princeton's psychology department, psychology.princeton.edu/news-events/.... The talk (and maybe the workshop) will be live-streamed for those interested in attending remotely.

I think it’s fair to argue that conflating inference and decision is the root of the majority of confusions in statistics. And arguably the original sin of frequentist methods.

Cool. Cool, cool.

I think it’s fair to argue that conflating inference and decision is the root of the majority of confusions in statistics. And arguably the original sin of frequentist methods.

Free visualization idea: Use component-wise, discretized Brownian bridges interpolated with cubic polynomials to animate leaves falling to form arbitrary patterns on the ground.

Free visualization idea: Use component-wise, discretized Brownian bridges interpolated with cubic polynomials to animate leaves falling to form arbitrary patterns on the ground.

Marg ̶i̶n̶a̶l̶arita Probability = 1

There is an absolute banger of a case study on modeling a linguistic experiment on Michael's patreon, thinking carefully through the whole data generating process and I am so looking forward to the video run through. Absolutely worth supporting him (do that and we'll get more :-))!

Just posted a new, long case study as well as a swag give away over on patreon dot com, www.patreon.com/c/betanalpha, in case you've been thinking about supporting.

Strongly support pronouncing "\" and "/" as "back pew" and "forward pew", respectively. Escape character hell immediately becomes much more fun.

An estimator penalty almost always defines a valid Bayesian prior model, but not always a productive one. A Bayesian prior model almost always defines a valid estimator penalty, but not always a productive one. Just because you can doesn't mean that you should.

Just posted a new, long case study as well as a swag give away over on patreon dot com, www.patreon.com/c/betanalpha, in case you've been thinking about supporting.

Strongly support pronouncing "\" and "/" as "back pew" and "forward pew", respectively. Escape character hell immediately becomes much more fun.

An estimator penalty almost always defines a valid Bayesian prior model, but not always a productive one. A Bayesian prior model almost always defines a valid estimator penalty, but not always a productive one. Just because you can doesn't mean that you should.

A tiny glimmer of joy.

You’ve build a probabilistic model and, for a small data set, computed posterior inferences with Hamiltonian Monte Carlo. Everything works great and, flush with confidence, you throw the model against all of your data only for everything to go to hell....

You’ve build a probabilistic model and, for a small data set, computed posterior inferences with Hamiltonian Monte Carlo. Everything works great and, flush with confidence, you throw the model against all of your data only for everything to go to hell....

I just realized that the reason why those huge Del Taco orders for large groups in undergrad were so good was that the tacos effectively steamed in the packaging, mimicking tacos de canasta.

Science isn’t computing a p-value. Science is working out when a p-value communicates meaningful information. . . . And how to better communicate information when a p-value doesn’t.

Science isn’t computing a p-value. Science is working out when a p-value communicates meaningful information. . . . And how to better communicate information when a p-value doesn’t.

Did you miss any of my writing last year? In addition to my big chapters I also released a few shorter pieces on the exploded logit model, symmetric prior models for positive spaces, and more.

Did you miss any of my writing last year? In addition to my big chapters I also released a few shorter pieces on the exploded logit model, symmetric prior models for positive spaces, and more.

A full pizza has fallen in the street and is constantly being run over. This has to be a metaphor for something.

A full pizza has fallen in the street and is constantly being run over. This has to be a metaphor for something.

Should initial conditions be forgot, and never brought to means? Should initial conditions be forgot, for con ver hence. For con ver gence my dear, for con ver gence, we’ll drink a coffee while we wait for con ver gence.

I’ve been looking into the details of a software called “Linger” and have not been able to get the Cranberries out of my head for days.

The horrors persist, but so does my writing. I reviewed all of my writing this year over on Patreon, www.patreon.com/posts/146964..., including a few pieces that until now have been Patreon exclusives.

'In particular, we firmly recommend: "Always write down the joint probability over all the settings you are interested in."' Many are saying!

The horrors persist, but so does my writing. I reviewed all of my writing this year over on Patreon, www.patreon.com/posts/146964..., including a few pieces that until now have been Patreon exclusives.

Naive averages failing once again.

I was looking into Voronoi diagrams for a side project, but I just could not follow the details. So I do want I am wont to do and dove into the details, working out a lot of geometry and making too many figures. HTML: betanalpha.github.io/assets/chapt... PDF: betanalpha.github.io/assets/chapt...

I was looking into Voronoi diagrams for a side project, but I just could not follow the details. So I do want I am wont to do and dove into the details, working out a lot of geometry and making too many figures. HTML: betanalpha.github.io/assets/chapt... PDF: betanalpha.github.io/assets/chapt...

Round and round it goes.

Round and round it goes.

If you incorporate the differences from weekdays and weekends as well (using a Dirichlet prior) you get down to 16 for a 50-50 chance: seehttps://drive.google.com/file/d/1ShvK_gHZOwOmtrf7Yy2ogi38YQaQJz_O/view.

Uncertainty quantification in business.

If you find yourself in need of distraction during the holiday season then consider dropping by the narratively generative modeling Discord server, www.patreon.com/posts/genera..., where we're discussing all aspects of modeling, implementation, and deployment.

Consumer PSA: A lot of stores will advertise an Xmas tree, but the package actually *contains* a Ymas tree. DON'T BE FOOLED.

Just sayin', turtle doves are gonna have to pick a side when the shit goes down between turtles and doves.

If anyone regrets not being able to attend my recent regression modeling course you can still access the slides, exercises, and recordings with a late donation. For donation instructions see betanalpha.github.io/courses/.

If anyone regrets not being able to attend my recent regression modeling course you can still access the slides, exercises, and recordings with a late donation. For donation instructions see betanalpha.github.io/courses/.

Shattered possibilities.