Software
My newest stuff is mostly here on GitHub, like my ChatGPT toolset (see also below). One of the most interesting things that is probably the one most used by others is probably the GrokConstructor written in Scala. I've also contributed a lot to the various Composum projects of my employer, IST GmbH Dresden, like the Composum AI and the Composum Nodes JCR Browser.
There is the mini-doodle-clone-with-a-twist GroupChoices, but here is also some older software I've written for fun or related to projects I've done: a Sudoku solver, an encoder that helps to memorize numbers with the master system, an error protection program CDProtect for CD-Roms, the C++ interface ObjCal to the CAL-BDD library, my BDD based planning program BDDPlan, Inti-gmp, an arbitrary length integer class for the programming language Sather, and some useful Shell Scripts for Linux.
If you are interested, here are also my questions and answers on stackoverflow:
ChatGPT / LLM based AI
Please look at the AI page for a more detailed listing.
ChatGPT tools contains a (as of 2023) fast growing toolset for me to work with ChatGPT. What I'm heavily working on and heavily using is my Co-Developer GPT engine which is a ChatGPT Plugin / server for a GPT you can start locally in a directory and use to read, search and modify the files in this direcory, access the web and trigger local actions (e.g. executing a build), all from within a ChatGPT chat. If you are interested in using ChatGPT from the command line or as IntelliJ external tool for things like creating test classes, that could be helpful for you. Also it contains some bookmarklets which you can grab from it's GitHub pages site. For instance there is one that I'm heavily using myself: it can grab the text from any page and have ChatGPT work on it - summarize it, answer questions for you etc. Check out also my fun little app ChatGPT chats with itself where you have two instances of ChatGPT with a different system prompt (that is, general character and / or goals) chatting with each other. That can be entertaining, interesting, and also productive in some cases.
Machine learning stuff
Simple neural network learning contains an explorational implementation of learning in neural networks, as well as some simple helper classes (automatic differentation, statistics, timing). It covers the RProp algoritm and an experiment to extend the gradient descent idea with using the second derivation to automatically select a good step width.
Machine learning explorations is a playground for me to explore implementations of machine learning techniques. In addition to basic stuff such as automatic differentiation it contains the CGP evolutionary learn algorithm, several implementations of simple backpropagation neural networks, and a partial implementation of running such calculations on the GPU with AparAPI.
GrokConstructor
The GrokConstructor, deployed at https://grokconstructor.appspot.com/, is a tool for semiautomatic construction and debugging of a special kind of regular expressions, that is used by the nifty Elastic Stack to parse logfiles. I wrote that in Scala, which is an amazingly well designed and productive language that combines functional and object oriented programming. When I write programs in Scala, I'm often amazed that an complex algorithm, which I would have been prepared to spend some debugging time on, works right out of the box. This is mostly since functional programming is way closer to the human thought processes than iterative algorithms, and because of Scala's powerful library for that.
GroupChoices
GroupChoices, is a little exploration which was created mostly in a Hackathon at T-Systems Multimedia Solutions. This is a tool that supports making choices in a group by allowing all members of a group to specify ratings for all possible choices, without needing registration and trying to capture your email address.
Caution: This was done mostly in the spirit "let's hack around with Scala and Twitter Bootstrap". So the source is more or less "write only code", though I maintain to an Scala adept it is still more readable than most even well written Java Code, since it is much more concise and to the point.
CDProtect
While CD error correction is quite good, you probably also have a couple of CDRs where some files are no longer readable. In many such cases the whole CD is worthless. As a remedy, CDProtect adds another layer of error protection to a CD that ranges over the whole CD. It operates in two modi. First, you can create a separate error protection file for an existing CD. Second, you can fill up a, say, 600MB Image with error protection information to the maximum 700MB that fit on the CD, and burn this. If an error develops later, you can usually reconstruct the CD.
The first mode already works for CDs (though it would currently take too much memory for DVDs). Unfortunately, I don't have enough time to carry this through. But fortunately, in the meantime there is DVDisaster around, which seems already to work good. So I think I won't continue this project. If you care, you can get the source here: errorprotect-0.1.tgz. It needs the library ecc-1.3.5.tgz for the actual ECC-correction. Of course, no warranties and so forth. ;-)
ObjCal
An object oriented C++ interface to the CAL BDD (Binary Decision Diagram) library. (BDDs are usually very compact representations for boolean functions, that can be manipulated efficiently.) The CAL library provides a rather clumsy syntax, and with ObjCal you can write more elegant code like:
Bdd b1("a"), b2("b"), b3("c"); // BDDs for variables a,b,c
Download objcal-0.1.tar.bz2 tested
under Suse Linux 8.2. Tested with version 2.1 of the CAL library.
Bdd res = -(b1 * b2) + b3; // BDD for not(a / b) / c
cout << res; // yields CNF-representation "abc + a*-b + -a"
BDDPlan
This is the C++-implementation of my PhD-Thesis, Planning with Binary Decision Diagrams. It solves planning problems specified in a fragment of the Planning Domain Description Language by means of BDDs (see ObjCal). The Planner works on planning problems specified in the ADL-fragment of PDDL (the Planning Domain Definition Language).
You can find more information on a separate page. Download bddplan-0.1.tar.bz2, tested under Suse Linux 8.2.
Inti-gmp
A replacement for the arbitrary length integer class INTI of the OO-programming language Sather. I liked Sather very much, since the language was designed to allow better software quality. It contains constructs like inheritable preconditions and postconditions for methods and class invariants, that are (optionally) checked automatically at runtime. This class was written for my study thesis Berechnung der Feigenbaumkonstante.
Download on a separate page.
Shellscripts
This is a collection of some shell- or PERL-scripts I wrote that I found useful and like to share. You can download the whole collection hpsscripts-0.1.tar.gz.
-
aless displays textfiles, Postscript-files, PDF-files, Unix-Manualfiles and the contenttables of tar-files while transparently uncompressing / / decrypting / processing these files as needed. It uses the scripts dviless, pdfless.
-
hardlndir replicates a directory (
hardlndir fromdir todir
) recursively using minimal space: all files in fromdir are hardlinked to the corresponding file in todir. I use this as a simple way to create incremental backups with rsync: I replicate the directory to my backup filesystem withrsync -avxHS dir backup/firstbackup
, and for the successive backups I dohardlndir backup/lastbackup backup/nextbackup
andrsync -avxHS dir backup/nextbackup
. Thus, only the changed files take additional space in the next backup; the unchanged files stay hardlinked to the last version. -
printpath displays all programs in the PATH whose names match the given argument. The argument can be a substring or a regexp.
-
findbiggest, findnewest and lessnewest are small useful one-liners. They find the newest / biggest files below the current directory or display the newest files with less. See the comments in the files.