Correlations

Correlations are ubiquitous. For example, news articles reporting that a research paper found no correlation between X and Y. Also, it is related to (in)dependence, which plays an important role in linear regression. This post will explain the Pearson correlation coefficient. The explanation is mainly based on the book by Hogg et al. (2018).

In the context of a book on mathematical statistics, certain variable names make sense. However, in this post, some variable names are changed to make the information more coherent. One convention which is adhered to is that single values are lowercase, and multiple values are capitalized. Furthermore, since in most empirical research we only need discrete statistics, the continuous versions of formulas are omitted.

2020-01-16

Benefits of writing blog posts

The first step into creating good habits is figuring out why exactly you want the habit. To me, writing blog posts seems like a good habit, but I'm unsure why. This post will attempt to convince the reader and myself of the benefits. I have combined my own ideas with the ideas by Terry Tao [1] and Gregory Gunderson [2], and grouped them.

Pedagogic benefits

Writing detailed expository notes is a way to practise research. This allows you to break free from the methods you are used to [2].
One can practise writing [1].
Writing allows one to test understanding of an idea [1]. It forces you to explain it clearly without hand waving. When aiming your text at colleagues or future employers, you cannot use jargon to hide your lack of knowledge [2].
Writing allows figuring out what exactly you do not understand or what you need to learn first [2].
Writing aids in structuring knowledge [2].

2019-12-29

Statistical power from scratch

In the 1970s the American government wanted to save fuel by allowing drivers to turn right at a red light (Reinhart, 2020). Many studies found that this Right-Turn-On-Red (RTOR) change caused more accidents. Unfortunately, these studies concluded that the results were not statistically significant. Only years later, when combining the data, it was found that the changes were significant (Preusser et al., 1982). Statisticians nowadays solve these kind of problems by considering not only significance, but also power. This post aims to explain and demonstrate power from scratch. Specifically, data is simulated to show power and the necessary concepts underlying power. The underlying concepts are the

2019-12-03

Niceties in the Julia programming language

In general I'm quite amazed by the Julia programming language. This blog post aims to be a demonstration of its niceties. The post targets readers who have programming experience. To aid in the rest of the examples we define a struct and its instantiation in a variable.

struct MyStruct
  a::Number
  b::Number
end

structs = [MyStruct(1, 2), MyStruct(3, 4)]

Functions and methods

For object-oriented programmers the distinction between a function and a method is simple. If it is inside a class it is a method, otherwise it is a function. In Julia we can use function overloading. This means that the types of the input parameters, or signatures, are used to determine what should be called. In Julia these are called methods. For example we can define the following methods for the function f.

2019-12-01

NixOS configuration highlights

I have recently started paying attention to the time spent on fine-tuning my Linux installation. My conclusion is that it is a lot. Compared to Windows, I save lots of time by using package managers to install software. Still, I'm pretty sure that most readers spend (too) much time with the package managers as well. This can be observed from the fact that most people know the basic apt-get commands by heart.

At the time of writing, I'm happily running and tweaking NixOS for a few months. NixOS allows me to define my operating system state in one configuration file configuration.nix. I have been pedantic and will try to avoid configuring my system outside the configuration. This way, multiple computers are in the same state and I can set-up new computers in no time. The only manual steps have been NixOS installation, Firefox configuration, user password configuration and applying a system specific configuration. (The latter involves creating two symlinks.)

2019-10-29

Entr

Having a compile and run shortcut seems like the most basic requirement for a developer. Most fully fledged IDE's therefore include it. For example, PyCharm will automatically detect the main file for the current Python project and run it with the correct virtual environment. This is all nice and well until it does not work out of the box. For example, when working with LaTeX most text editors can install a plugin which introduces a compile and run shortcut. Or, if you are lucky you can write down some script at some place in the editor which will execute upon a certain key press. This works as long as your static script is able to infer the correct file to execute. If not then the editor command needs to be changed for each project.

2015-08-25

QoS setup using Tomato in combination with an Experiabox

This is a copy of my blog post at Blogspot. It is mostly here for myself, so that I can compare my writing in 2015 with newer writings.

Introduction

In this blog my setup for QoS (Quality of Service) will be explained. The QoS is used in a home network with five users. The home network has an maximum download/upload speed of 650 / 35 kB/s. The QoS had to be introduced because of 'slow internet' noticed by the users while browsing websites, gaming or using VoIP. This problem was growing ever year because of more clients (i.e. smartphones, tablets and laptops) and data usage (cloud backup, HD stream availability). The key in solving the problem is to avoid using too much bandwidth. When too much bandwidth is used the data will pile up resulting in slow internet packet delivery. To limit the bandwidth it is important to slow down the big users like streaming and cloud backup. Last year a solution using the program cFosSpeed was implemented. This program ran on all the Windows devices and limited data based on the responsible process. Unfortunately the program could not run on android, meaning that those devices weren't limited at all. This rendered the solution completely useless. The solution now used is based on a router with some advanced firmware. The router knows nothing about the responsible processes, but is only looking at the packets. This results in a completely platform independent system which works without any set-up at the client side.