Analysing Net Benefit Analysis
This post is an exploration of the “NetBenefit curves” of Vickers and Elkin, also promoted by Steyerberg. I run regressions on some dataset, and investigate ...
Posts by Year
2024
Standard Python Logging with the Joblib Loky Backend
The main issue with python logging and joblib multiprocessing is that the loky backend on joblib does not offer a way to set up the worker processes where th...
Hosting things on my Synology NAS
I use the domain elhult.duckdns.org to host things for private use. It points to my NAS, a DS214+ running DSM 7.1 from Synology. It is useful for backing up ...
A proof for the back-door criterion
A very common way to quantify a causal effect is inverse-probability-weighting (IPW) with back door adjustment. You can find the method described in many pla...
Watching the computer do the Simplex algorithm
Some time ago, I needed a refresher on linear programming. I observed, among other things, the sparsity of solutions. It felt like some nice intuition about...
The F-statistic you see in OLS output
[\def\E{{\mathbb E}} \def\R{{\mathbb R}} \def\betaTrue{{\beta_{\circ}}} \def\eye{I} \def\coloneqq{:=} \def\eqqcolon{=:} \def\rank{\operato...
2023
Real World Data Sets are Messy: Case Study Elec2
Benchmark data sets that are ill documented provide harm to the ML community and the research output, even if they are “real world”. Recent work (link) put t...
Functools Partial
Partial application We have a function f(a,b) but we want to fix the values of b some value, and get a new function of a only. In pseudocode, we want to sa...
Running means and covariances
Welfords algorithm
Documenting correctness in Haskell
In the Functional Programming 1 course we teach at Uppsala University, there is some confusion about how to document code. Every year, there is confusion. I ...
Quantile minimization should give good machine learning, but is hard. Does TERM solve that?
When optimizing a machine learning model for an optimal parameter \(\hat\theta\), we often rely on the empirical risk minimization framework,
How fast get you read the BNO055?
I am trying to collect some high frequency accelerometer data for a machine learning hobby project. Since I had the Adafruit BNO055 board lying around, that ...
2022
How to Log, with Scikit-Learn
I recently dug deep in to logging and scikit-learn. It lead to a SO question, and a few hours later, noone but me has attempted a solution. I thought I would...
Nonparametric prediction intervals
There is a simple formula for a onesided nonparametric prediction interval under exchangeable data. One can derive it very directly under iid assumptions, an...
Variational Inference - An Example Calculation
I was working though problems from the compendium Pattern Recognition - Fundamental Theory and Exercise Problems by Arne Leijon and Gustav Eje Henter. One of...
Recording Audio for Machine Learning with PyAudio
In a PhD course I am taking, we use Hidden Markov Models for speech recognition. To make it more exciting, we are encouraged to collect our own training and ...
Classical OLS
[\def\E{{\mathbb E}} \def\R{{\mathbb R}} \def\betaTrue{{\beta_{\circ}}} \def\eye{I} \def\coloneqq{:=} \def\eqqcolon{=:} \def\rank{\operato...
Empirical Bernstein Bounds
In machine learning, we often rely on the possibility to estimate an expectation by a sample mean. One example is the hope that a model selected by empirica...
Lebesgue integral, Radon-Nikodym, Mutual Information and relative measurability
This post continues the previous example. It adds in definitions and terms relating to Lebesgue integrals and entropy.
Probability theory definitions by example
This post is a note-for-memory regarding definitions in probability theory, and how they relate, by giving a simple example.
CVaR, VaR and quantiles
There are funny concept in mathematical finance, that have strange abbreviations and funny names. It seems part of the problem is that one tries to capture ...
2021
Example of limits and Lebesgue integration
Examples from Lebesgue theory and examples in integration and taking limits. Specifically, bounded convergence.
On Haskell Builds
Currently, Im about to finish my AoC 2016 code in Haskell. The main motivation is that I will be Teaching Assistant in a Functional Programming course this f...
Jekyll Minial Mistakes
Today I migrated to Minimal-Mistakes, quite possiblu the most common Jekyll theme for personal blogs in the ML-AI clique. I realized it was worth it to migra...
The independent complement of a random variable
This post came to be when I was reading arxiv:2008.01883. They use some definitions and terms from probability theory that I needed to revise, and in the pro...
Reflections on the Bregman Divergence
Reading arXiv:2008.01883v2 I was thinking a little about the Bregman Divergence. It is a measure of distance sometimes used in probability theory that I am n...
Recording videos as a TA
As a PhD student, I am often a teaching assistant (TA) in the courses tought at my department. With the covid-19-pandemic as a driver, we have more and more ...
An invertible BatchNormalizer1D
In Normalizing Flows, you might want to normalize the data at some point. Either in the input layer, because you want your model to learn the data normaliza...
Installation project: New radio
I am writing this as a note for memory regarind the installation of a new head unit in my Citroen C3 from 2002. I write as I research, so I’m by far no exper...
Some standard tricks for LaTeX
There is a bunch of things I have repeatedly been looking up over the last months. Here they are - as a kind of easy-to-find repository of snippets.
My current LaTeX setup
tl;dr LaTeX is a constant source of frustration. This setup relieved some of mine:
Some nix-trix
So today I tried to run some stuff on a Ubuntu server from my Windows 10 machine. I did it in four steps that I liked, and that I wanted to write down for fu...
Some exploration of conda
I’ve spent almost a full day battling conda. My goal was to have a simple environment.yml file that I could conda env create --file environment.yml both on m...
Update equations for Recursive Least Squares
The Recursive Least Squares (RLS) method is commonly used, but I did not find any derivation of it that suited me. Since I did the derivation for my own sak...
GitHub Pages with MathJax
So I thought I’d write a line about how it is to have a first experience in using GitHub Pages and enabling MathJax. I had not used Jekyll at all before, so ...