Put this on a server, turn off the network logging and you have an anonymous live chat without database, cookies or even javascript.

Currently it contains about 47,000 (small) images in about 30 categories. Still some way to go...

Continuation of frontpage-archive-2

Continuation of teletext-archive-unicode

A collection of code that aggregates my various archive repos.

Recently spent all neural net coding in this repository. Sometimes i browse arxiv.org, and try to implement a technique from the next paper. And, of course, i have my own artistic desires.

The idea is basically to write or extend a drawing application with a "magic pen" that allows configuration through a text prompt. E.g. "pixelart mid-age mossy stone" and then draw on an existing graphic to adjust the details. Work is currently superseded by the CLIPig2 implementation in nn-experiments.

Some experiments i wanted to save. Superseded by nn-experiments!

An abundance of S3M files from my early days.

A web page that shows teletext pages through history!

I don't watch TV, i just collect the programme for statistical reasons.

A generic tool for all the archive repositories. Makes it easy to iterate through all commits and the individual file data.

Continuation of teletext-archive-unicode

Continuation of frontpage-archive-2

Unfortunately, the site went down. It was hosting little programs that fight each other in a little square grid. I developed a local simulator and designed the number one battle bot:

I really liked teletext as a kid. It was looking like computer stuff. Now i'm just scraping it regularily to build some news trend analysis. Well, news sounds so regular. Let's say brainwashing dumbf**k that we get poured over our heads from day to night.

I was following some crazy stream of desire to once and for all point the finger at the true enemies of beauty, creativity and well-being of nature and mankind... using the nasdaq.com API. Not the official one of course, but their internal website API which is not restricted by quota. Still it was days and days of downloading to get (probably) the whole graph of shareholders and insiders from the nasdaq database.

Can i point the finger now? Sure! Blackrock, Vanguard, Fidelity, T. Rowe Price, Morgan Stanley, Wellington, Bank of America ... simply the big bubbles.

It is actually super creepy! The only fun fact i got from this experiments is this: Amazon is a little kitten compared to those investment funds. But thinking seriously about it: If Amazon already dictates prices and work conditions how f**king mighty is Blackrock?

Same purpose as the teletext-archive-unicode - analyzing trends in news. Here the titles, teasers and image texts of a number of different german papers are stored.

Learned: That committing a hundred html pages every 15 minutes makes the repository grow to over 700mb in two weeks. That's what happened to the predecessor. And i didn't really wanted to wait for github staff to examine and close the repo. The predecessor still collects a snapshot every month. Just to be able to take a look at the raw data and measure the development of applied scripts, ads and tracking.

My first attempt to use the github actions for scraping data.

The KEK is an organisation collecting the shareholder relations of all newspapers, radio and tv stations a.s.o in germany. This repo stores a snapshot of their database every day. So in fact, it actually collects the amount of work that the KEK is spending every day to update their database.

I was trying to write my own github trending that also finds good repos that do not have stars and attention yet. It's not super easy to process all the github events that exist. In the course of trying i came across some interesting commit messages. good-github collects them using a hand-crafted list of words and publishes the messages in a daily journal.

Let's see ... if i need this often enough i might actually develop a good thing there..

A collection of github repos which have been pushed to in some automatic fashion during 2018: defgsus.github.io/gharchive-stats/automatic.html

Learned: some patience

You are currently viewing it's github pages!

Just stumbled over the Web Audio API. Actually i wanted to see if it's possible to program an efficient synthesizer in javascript. As it turns out, this API provides an interface that is exactly for building synthesizers...

So then i wanted to see if React can be used to drive the event-loop of the step-sequencers. Did not want to use redux this time and learned about React Hooks and especially useReducer and all these magic functions.

As to be expected, the timing of setTimeout driving the react dispatchers is not good enough to drive an audio machine correctly (where 20 milliseconds off are already perceived as not in the beat). It's fun though! Have a try...

The data output of office-schedule-scraper.

One of the important things each day is to generate data for this repository ;)

This is recording hundreds of public office calendars across germany at an interval of 15 minutes.

It's much more work than i anticipated in the first place and i'm still not sure exactly for what purpose.

The web has become quite an amazing place: Train a model on some of it's data. Ask for a depiction of your favorite fantasy book. And it just does!

This is a somewhat usable tool for generating images using the CLIP critique. Check the walkthrough and some experimental results.

Actually, this is just a small dev tool to download images (porn images mainly), rate them via CLIP and search through them using natural language and reference images.

Guess the virus has taken it's toll. Or maybe because i'm 40. Anyways, this is my programming blog. Using Jekyll and Jupyter notebooks as the basis for static site generation.

Spy on your own browser actions (request, keystrokes, mouse actions) using this small web extension and Elasticsearch.

Learned about the (somewhat unified) Browser Extensions API (MDN, Chrome) which is so powerful that i do not trust any browser extension anymore, except my own ones, and uBlock Origin and uMatrix. The latter, though, i just trust out of good faith.

Thought: If i write the raytracer (in this case a sphere-tracer) in python, there are endless pythonian things i can do on top to create amazing images.

Learned: Writing a raytracer in C or Cuda instead of python is not premature optimization. The python raytracer is so terribly slow that i actually did not fix the immediate bugs and just stopped the experiment.

This is a library for Elasticsearch queries. They provide their own library for that but it does not help with either generating the right query or accessing the results in a sane manner. Elastipy has a fully typed interface and tries to be smart with aggregation responses, converting them into simplified dicts or pandas DataFrames.

Learned: Generating documentation and examples from jupyter notebooks. And that the whole documentation system in python is good only on the surface. Actually it's painful, not well documented and time consuming. No surprise, why all the major libs have their own documentation infrastructure (hacked Sphinxes).

Tried: A 2d minecraft.

Learned: That in order to pass objects, you have to completely destroy them or build a tunnel or somehting..

This came out of a deep depression through covid winter 2020 with too much home schooling and home officing and no place to hide. In the end i did not came across the challange of designing a physical-model-based character that actually can walk. Tried evolutionary algos with complicated fitness evaluation and training pools stored in elasticsearch.

Started reworking the pickpack idea into python to experiment with Maximum Expected Utility described by Artificial Intelligence: A Modern Approach (Russel & Norvig). But it's still missing a straight game design.

Basically, you are picking and packing in an amazon fulfillment center against a bunch of super-efficient robots. The idea is, that the robots would actually need to cheat (stay in your way, put wrong stuff into your parcels, ..) if you are really good. Of course, if the ever-present fulfillment center's performance measures are too low, the manager should come complaining.

Learned: In python you won't get millions of search branches checked per second so you have to rely more on heuristics and cheats.

It was hosted on kaggle and a lot of teams joined. I only used relevance vector regression on the small MRT-based CSV tables which bought an evaluation error of 16.4% and place 891. The top solutions where 15.6%.

Learned: That this particular problem in the competition was not helped by mere compute. The best teams used all kinds of complicated things and ensembled them together to gain just one percent better than a simple RVR on a fraction of the data.

The daily export of parking-scraper.

It scrapes all those different little websites where free parking lots are displayed and stores the numbers - every 5 minutes. There's a small article about the 1st year of data.

Actually it's a political statement. Or it was planned to be. But it's missing the game idea. See pypickpack

Another personal plaything related to shadertoy.com. Besids the official API it also allows crawling the shadertoy website / internal API to retrieve comments.

One of the goals is to be able to somehow search for specific formulas / syntax trees, by parsing all the GLSL in python. The parser somewhat works by adjusting an old Backus-Naur for C-- that i found and adding the few GLSL specific things (glsl.lark) and run it in lark. But it exponentially explodes in runtime when shown long and complicated formulas.

Guess everybody has to release one of these libs once. It's not particularily good or clever, or even tested twice.

Just some experiments in the realm of micro services.

Quite helpful for a couple of work projects. Not a mature generic lib, though.

This just good useful to be able to install via pip when moving database content via manage.py dumpdata.

Inspired by the original Botwars for which i don't find the link right now. But this seems to be a somewhat worthy predecessor. Basically it's about playing a competitive game by providing source code. The original Botwars was simply a grid of memory cells which are executed and which try to overwrite and patch each other. The winner is the one with the most occupied cells.

My version is more the classic 2D environment, using signed distance functions which allows for easy implementation of distance sensors and the like.

If you rent a second-hand server, e.g. one that has been used for something before, you'll most likely get a ton of requests of content that was there before, as well as malicious attacks to all kind of ports and, of course, common PHP urls. This project is a django that collects the content of all the log files, stores all received POST requests, collects the whois of IPs and also presents a honeypot username/password form.

There was a time, when we used discord for work meetings. This little bot joined. It would post Live of Brian or Bladerunner citations with the person names replaced by the names of the chat members. It could display tickets from YouTrack. And it would engage in a german Eliza therapy chat.

Just a plaything with python generators and mathematical note generation. It has some good starting point but is not well suited to manage chords or polyphone sequences.

OpenGL rendering graph and A* search movement.

The rendering is particularily nice, optically. It uses sphere tracing through a pre-generated 3D distance field for the soft-shadows.

Also started an editor in PyQt. Learned: Programming Qt in python is almost fun, compared to C++.

At the time i simply found it amazing that one can freely post javascript on blogspot.com. So i published a few data-mining articles. When the data became too much for one post (the posting interface would not work anymore) i moved it to this repo and requested it externally.

A nasty parent tool to overview what the kids are actually doing on the web. Three reasons why i did not continue for long:

1. They got old enough to truly deserve a private sphere.
2. All of them got mobile data contracts at some point.
3. This logger tool mostly collected HD video data streams :D

Someone leaked the chat logs of an internal AfD whatsapp group. I did some word statistics.

This was my fulltime work between 2013 and 2016. An audio-visual fulldome production and live software with a lot of nifty features.

Actually quite proud of it (except the name). No good performance but a helpful API for the usual vector-in-space manipulation.

Coming from C++ and playing with yacc/bison i was fascinated by python's real-time everything. To be able to create a compiler during runtime was a new thought to me.

From the good ol' Shadertoy times.

There is this user called Kali who posted on fractal forums this amazing formula. It got very popular on Shadertoy. And i used it fulldome once in a while.

This is a C++ implementation.

Well, it was just one class where a friend was teaching at the time.

Wrote it to get a grip on alpha beta tree search

A project with Francis Monkman. Through friends in Russia he got into contact with this game called Shatra. It's the mongolian version of a checkers-like game. Some russian suggested to use chess figures which makes it a kind of turbo-checker-chess.

It's a cool game and Francis has been working for decades on efficient alpha-beta pruning for this particular rule set.

We forked ChessX which is a Qt/C++ project, mainly for archiving and studying chess games, and extended it to both Shatra versions and added a play mode.