Un garçon pas comme les autres (Bayes)

A blog about statistics, I guess.

Random C++ Part 2: Sparse partial inverses in Eigen

Stan
Sparse matrices
Autodiff
Eigen
Sometimes you just need some elements of the inverse of a sparse matrix. Sometimes you’re working in C++. This is that time.
Sep 5, 2024
Dan Simpson

Random C++ Part 1: Building a block sparse matrix in Eigen

Stan
Sparse matrices
Autodiff
Eigen
I was writing a longer thing, but it was too long, so hey. Let’s just do this for a change
Sep 4, 2024
Dan Simpson

An unexpected detour into partially symbolic, sparsity-expoiting autodiff; or Lord won’t you buy me a Laplace approximation

JAX
Laplace approximation
Sparse matrices
Autodiff
Exploiting linearity and sparisty to speed up JAX Hessians and slowly ruin my life.
May 8, 2024
Dan Simpson

Diffusion models; or Yet another way to sample from an arbitrary distribution

Diffusion model
Introductions
If you gaze for long into an abyss, the abyss creates a cutting-edge AI method. So saddle up your Ganzfeld effect, girls. We ride at dawn.
Feb 9, 2023
Dan Simpson

Markovian Gaussian processes: A lot of theory and some practical stuff

Gaussian processes
Fundamentals
Theory
Deep Dives
Well this is gonna be technical. And yes, I’m going to define it three ways. Because that’s how comedy works.
Jan 21, 2023
Dan Simpson

Sparse matrices part 7a: Another shot at JAX-ing the Cholesky decomposition

JAX
Sparse matrices
Autodiff
I work in R a lot so I should be used to weird syntax. This part looks at the non-zero pattern.
Dec 2, 2022
Dan Simpson

MCMC with the wrong acceptance probability

Fundamentals
MCMC
Bayes
Sometimes I chat work with people. Sometimes an interesting factlet comes up. Sometimes I blog about it. This is one of those times.
Nov 23, 2022
Dan Simpson

On that example of Robins and Ritov; or A sleeping dog in harbor is safe, but that’s not what sleeping dogs are for

Fundamentals
Survey sampling
MRP
Bayes
Look, it’s a dull example of Bayes being bad. But it comes up often enough to be worth talking about. I’m going to, unsurprisingly, argue that Bayes isn’t bad. Neither are Robings/Ritov/Wasserman wrong. They’re just looking at the problem through a different lens.
Nov 15, 2022
Dan Simpson

Priors for the parameters in a Gaussian process

Prior distributions
Gaussian Processes
PC priors
If you’re not a machine learner (and sometimes if you are), Gaussian processes need priors on their parameters. Like everything else to do with Gaussian processes, this can be delicate. This post works through some options.
Sep 27, 2022
Dan Simpson

A first look at multilevel regression; or Everybody’s got something to hide except me and my macaques

Multilevel models
Visual diagnostics
Prior distributions
fundamentals
A small introduction to multilevel models. Why? Because I said so, that’s why. And you will simply not believe what happens to residual plots.
Sep 6, 2022
Dan Simpson

Priors part 4: Specifying priors that appropriately penalise complexity

Prior distributions
fundamentals
PC priors
Let us not lie. Specifying priors are hard. This post steps through a technique for setting priors that I think gives a good basis for realistically complex problems.
Sep 3, 2022
Dan Simpson

Tail stabilization of importance sampling etimators: A bit of theory

Importance sampling
Computation
Truncated importance sampling
Windsorized importance sampling
Pareto smoothed importance sampling
PSIS
Look. I had to do it so I wrote it out in detail. This is some of the convergence theory for truncated and winzorised importance sampling estimators
Jun 15, 2022
Dan Simpson

Sparse matrices 6: To catch a derivative, first you’ve got to think like a derivative

JAX
Sparse matrices
Autodiff
Open up the kennels, Kenneth. Mamma’s coming home tonight.
May 30, 2022
Dan Simpson

Sparse Matrices 5: I bind you Nancy

Sparse matrices
Sparse Cholesky factorisation
Python
JAX
A new JAX primitive? In this economy?
May 20, 2022
Dan Simpson

Sparse Matrices 4: Design is my passion

Sparse matrices
Sparse Cholesky factorisation
Python
JAX
Just some harmeless notes. Like the ones Judy Dench took in that movie.
May 16, 2022
Dan Simpson

Sparse Matrices 3: Failing at JAX

Sparse matrices
Sparse Cholesky factorisation
Python
JAX
Takes a long drag on cigarette. JAX? Where was he when I had my cancer?
May 14, 2022
Dan Simpson

Sparse Matrices 2: An invitation to a sparse Cholesky factorisation

Sparse matrices
Sparse Cholesky factorisation
Python
Come for the details, stay for the shitty Python, leave with disappointment. Not unlike the experience of dating me.
Mar 31, 2022
Dan Simpson

Sparse Matrices 1: The linear algebra of linear mixed effects models and their generalisations

Sparse matrices
Linear mixed models
Hubris. Just hubris. But before the fall comes the statement of purpose. This is that statement.
Mar 22, 2022
Dan Simpson

Barry Gibb came fourth in a Barry Gibb look alike contest (Repost)

Computation
Assessing algorithms
A repost from Andrew’s blog about comparing computational methods for performing a task. (Lightly edited.) Original posted 20 October, 2017.
Jan 26, 2022
Dan Simpson

Why won’t you cheat with me? (Repost)

Prior distributions
Fundamentals
Design dependence
A repost from Andrew’s blog about how design information infects multivariate priors. (Lightly edited. Well, a bit more than lightly because the last version didn’t fully make sense. But whatever. Blogs, eh.) Original posted 5 November, 2017.
Dec 9, 2021
Dan Simpson, Dan Simpson

The king must die (repost)

Bayesian Lasso
Variable Selection
Fundamentals
Theory
Things that don't work
A repost (with edits, revisions, and footnotes) from Andrew's blog about how much I hate the Bayesian Lasso. Originally published 2nd November, 2017.
`Dec 8, 2021`{=html}
Dan Simpson, Dan Simpson

Getting into the subspace; or what happens when you approximate a Gaussian process

Gaussian processes
Fundamentals
Theory
Fuck man, I don’t know about this one. A lot of stuff happens. At some point there’s a lot of PDEs. There are proofs. Back away.
Nov 24, 2021
Dan Simpson

Yes but what is a Gaussian process? or, Once, twice, three times a definition; or A descent into madness

Gaussian processes
Fundamentals
Theory
Deep dive
Gaussian processes. As narrated by an increasingly deranged man during a day of torrential rain.
Nov 3, 2021
Dan Simpson

Priors: Fire With Fire (Track 3)

Prior distributions
Fundamentals
Objective priors? In finite dimensions? A confidence trick? Yes.
Oct 17, 2021
Dan Simpson

Priors: Whole New Way (Track 2)

Prior distributions
Fundamentals
Conjugate Priors? The crystal deoderant of Bayesian statistics
Oct 16, 2021
Dan Simpson

Priors: Night work (Track 1)

Prior distributions
Fundamentals
Priors? Defined. Questions? Outlined. Purpose? Declared.
Oct 15, 2021
Dan Simpson

\((n-1)\)-sane in the membrane

Teaching
Fundamentals
Opinionated
Windmills? Tilted. Topic? Boring. (n-1)? No.
Oct 14, 2021
Dan Simpson
No matching items