Click on the image below for a video demo.
An experimental typing tutor in alpha stage. Implements typing tutor that generates random characters in a way that maximizes a chance for a mistake to be made. In effect, this means that the weakest character combinations have the highest chance to be in the practice sentence. The driving idea for this project is the fact that different character combinations make our fingers more prone to subsequent mistakes hence practicing every possible pair of characters should improve typing speed and decrease mistakes.
The underlying model is based on Markov chain where probabilities are the ratio of correct and wrong hits of character pairs. In addition to this, there is a pressure applied from the speed of typing meaning that even if the character is typed correctly but the speed was lower that average it would still be counted as wrong. To further increase (or decrease) the pressure, the average typing speed is smoothened with exponential moving average, meaning that only fairly recent averages will be counted (the better you get the harder it becomes). This also means that if you have a bad streak the pressure is decreased faster.
Currently only the development package is available for Arch Linux in AUR.
After installing you can start the executable from the command line:
markov-typing
You will probably need GCC 8 to compile the program. Clang 6 will also work but it has to use GCC 8 filesystem library (libstdc++). libc++ has filesystem support starting from Clang 7.
Build requirements:
- GCC 8.1
- Qt5
- CMake
- make or ninja
Build steps:
cmake -H. -Bbuild -DCMAKE_BUILD_TYPE=Release
cmake --build build
If you want to try building with compilers that support C++17 but don't have the filesystem library try the following:
cmake -H. -Bbuild -DCMAKE_BUILD_TYPE=Release -DUSE_BOOST_FS=TRUE
cmake --build build
After building you can run the executable ./build/markov-typing. The matrix is
stored in JSON format either in $XDG_DATA_HOME/MarkovTyping or
$HOME/.local/share/MarkovTyping and is loaded from the same location so the
progress is saved. Since this is experimental so far, the same save file might
not work between different commits.
This is mostly an experiment to see how such tutor would work and anything can change and nothing can be considered stable.
If you play with it and have some feedback/ideas please let me know.