Programming Leftovers

posted by Roy Schestowitz on Oct 27, 2025

• Rlang ☛ Double Descent Explained

  This is a post about the famous double descent phenomenon in machine learning. I aim to constrct the simplest possible example of double descent and explain exactly how it arises and why it doesn’t contradict the bias-variance tradeoff.

• Daniel Lemire ☛ Thinking Clearly

  That’s a terrible idea. Even simple projects become challenging if you are ambitious. There is no need to complexify them. For example, seven years ago, we aimed to create a JSON parser faster than anything on the market—a simple idea. A senior colleague in computer science saw me at a campus coffee shop while I was working. He asked what I was busy doing… When I told him, “We’re writing a fast JSON parser,” he laughed. He would later admit that he thought I was joking: how could I work on something so mundane. I wasn’t kidding. The result was simdjson, a JSON parser four times faster than anything else at the time. By keeping our project conceptually simple, we made success easier.

  Complexity is a burden, not a badge of pride.

• Maury ☛ You already have a git server

  This is a great way to sync code between multiple computers or to work on server-side files without laggy typing or manual copying. If you want to publish your code, just point your web server at the git repo: [...]

• Alexandru Nedelcu ☛ Scala 3 / No Indent

  Scala 3 came with significant indentation (🏛️). And I still dislike it, despite trying hard. So I’m switching back to braces.

• Alexandru Nedelcu ☛ Scala 3 Significant Indentation Woes: Sample

  Here's a fairly straightforward Scala 3 sample, using significant indentation. Can you spot the compilation error?

• R / R-Script

  • Rlang ☛ Approximating evidence via bounded harmonic means (and HPD regions with known volumes)

    Following a suggestion by Christian Hennig at JSM 2024, I started working with my PhD student Dana Naderi on a detailed assessment of the method we proposed in 2009 with Darren Wraith for evidence approximation. [...]

• Python

  • Didier Stevens ☛ Update: dnsresolver.py Version 0.0.4

    If provided, the value will be interpreted as a Python function and called when there is a match. The function must inspect the request and update the reply. Arguments to the function are request, reply and dCommand.

  • Didier Stevens ☛ Quickpost: PEP 515 – Underscores in Numeric Literals

    I’m familiar with notations in Python like 0b10000001, but not with an underscore thrown in to make the number more readable.

  • Hackaday ☛ Making Math Less Stressful With A Python Super-Calculator

    In a recent write-up, [David Delony] explains how he built a Wolfram Mathematica-like engine with Python.

• Java/Golang

  • Daniel Lemire ☛ Flame Graphs in Go

    The hardest problem in software performance is often to understand your code and why it might be slow. One approach to this problem is called profiling. Profiling tries to count the time spent in the various functions of your program.