Dear colleagues,

Assessing the functionality of the libtglang.so library, please pay attention to the following aspects:

1. Training dataset is loaded from the "resources/groundtruth/1...99" folders on initial call of the "tglang_detect_programming_language()" function. As well as model training occurs on this stage. Therefore, please do not take into account the time it takes for the first execution to evaluate alg performance. Subsequent function calls do only inference according to the contest conditions.

2. Due to significant development time constraints the training dataset contains only 42 langs:). You can modify the training dataset without rebuilding the library by adding new files and folders to resources/groundtruth/1...99 (the folder number corresponds to the language number according to the "TglangLanguage" enum in "tglang.h").

3. If lib built in debug mode (#define DEBUG in src/libtglang/language_detection.hpp) it prints statistical data to graphic *.pgm files. These files are useful for analyzing the lib's operation and can explain its functioning principles.

Unfortunately, this submission has been disqualified after our preliminary tests for exceeding the 500ms limit per text sample.