HSL-Reviewer

Improve your shapes and intuition in the game of Go.
Tailered to your weaknesses from your own games. Using human supervised learning for questions and answers of your skill level.

Disclaimer

I made this in a few hours for myself and decided last minute that others might find it useful too. The code is a mix of lightvector's KataGo python scripts, mine and ChatGPT's so it's a bit of a mess. I can't guarantee that it works correctly or even works at all on your system. That said, I'd apreciate any suggestions, bug reports or contributions.

Prerequisites

1. KataGo with a model.
2. A HSL model. For example this model from lightvector. Check the Discord server for possibly newer versions.
3. Python. Version 3.10.6 works for me.
4. Run pip install -r requirements.txt to install the necessary python packages.
5. Install PyTorch.
6. Change the parameters in Reviewer.bat to your own. If you're not sure about the device, CPU is a slow but safe bet. Format (SGF and PLAYER are optional): [HSL_MODEL] [HSL_DEVICE] [KATAGO_EXE] [KATAGO_ANALYSIS_CFG] [KATAGO_MODEL] (SGF) (PLAYER)

Create flashcards

1. Run Reviewer.bat
2. Set the color you were in the game you want to review.
3. Optionally change the HSL sliders.
4. Drag & drop your game sgf.
5. Wait until the console says === REVIEW DONE ===

There are also some options you can tweak at the top in hsl_reviewer.py.

Train

1. Run Trainer.bat
2. Load a card and think of a move.
3. Click on the board to see the answer.
4. Delete the card if you feel the answer is easy.
5. Load the next card.

It should also work on linux and mac. Just run hsl_reviewer.py or hsl_trainer.py directly.

Screenshots

X = Game move
O = HSL move