KaTrain v1.0

This repository contains tool for analyzing and playing go with AI feedback from KataGo.

The original idea was to give immediate feedback on the many large mistakes we make in terms of inefficient moves, but has since grown to include a wide range of features.

Some uses include:

• Review your games to find the moves that were most costly in terms of points lost.
• Play against AI and get immediate feedback on mistakes with option to retry.
• Play against weakened versions of AI.
• Play against a stronger player and use the retry option instead of handicap stones.

Installation

Installation for Windows users

• Download the repository by clicking the green Clone or download on this page and Download zip. Extract the contents.
• Make sure you have a python installation, I will assume Anaconda (Python 3.7), available here.
• Open 'Anaconda prompt' from the start menu and navigate to where you extracted the zip file using the cd <folder> command.
• Execute the command pip install numpy kivy_deps.glew kivy_deps.sdl2 kivy_deps.gstreamer kivy
• Start the app by running python katrain.py in the directory where you downloaded the scripts. Note that the program can be slow to initialize the first time, due to kata's gpu tuning.

Installation for Linux/Mac users

• This assumed you have a working Python 3.6/3.7 installation as a default. If your default is python 2, use pip3/python3. Kivy currently does not have a release for Python 3.8.
• Git clone or download the repository.
• pip install -U kivy numpy
• Put your KataGo binary in the KataGo/ directory or change the engine.command field in config.json to your KataGo v1.3.5+ binary.
  • Compiled binaries and source code can be found here.
  • You will need to chmod +x katago your binary if you downloaded it.
  • Executables for Mac are not available, so compiling from source code is required there.
• Start the app by running python katrain.py. Note that the program can be slow to initialize the first time, due to KataGo's GPU tuning.

Manual

Play

• Human is simple play with potential feedback, but without auto-undo.
• Teach will give you instant feedback, and auto-undo bad moves to give you a second chance.
• AI will active the AI in the dropdown next to the buttons:
  • Default is full KataGo, above professional level.
  • Balance is KataGo occasionally making weaker moves, attempting to win by ~2 points.
  • Jigo is KataGo aggressively making weaker moves, attempting to win by 0.5 points.
  • Policy is the top move from the policy network (it's 'shape sense' without reading), should be around high dan level depending on the model used.
  • P+Pick will pick a pick_n + <number of legal moves> * pick_frac moves at random, and play the best move among them.
    • P+Local will pick such moves biased towards the last move with probability related to local_stddev.
    • P+Tenuki is biased in the opposite way as P+Local, using the same setting.
    • P+Influence is biased towards 4th+ line moves, with every line below that dividing both the chance of considering the move and the policy value by influence_weight. Consider setting pick_frac=1.0 to only affect the policy weight.
    • P+Territory is biased in the opposite way, towards 1-3rd line moves, using the same setting.
  • P+Noise mixes the policy with noise_strength Dirichlet noise. At noise_strength=0.9 play is near-random, while noise_strength=0.7 is still quite strong. Regardless, mistakes are typically strange can include senseless first-line moves.
  • <Pause> pauses AI moves, in case you want to do analysis without triggering moves, or simply hide the evaluation dots for this player.

Analysis

Keyboard shortcuts

Configuration

Configuration is stored in config.json. Most settings are available to edit in the program, but some are not. You can use python katrain.py your_config_file.json to use another config file instead.

The settings panel

The trainer block has the following options to tweak for engine assisted play and reviewing:

• eval_off_show_last: when the eval checkbox is off for a player, show coloured dots on the last this many moves regardless.
• undo_eval_threshold, undo_point_threshold: prompt player to undo if move is worse than this in terms of points AND evaluation.

The following options are relevant for the balance score AI play mode.

• balance_play_target_score: indicates how many points the AI aims to win by when using 'balance score'.
• balance_play_randomize_eval: when not needing to balance score, the AI will pick a random move which is at least this good as long as it stays ahead.
• balance_play_min_eval: when needing to balance score, the AI will pick a move which is at least this good.
• balance_play_min_visits: never pick a move with fewer playouts than this.

The cfg file has additional configuration for KataGo, which are documented there.

Configuring feedback

• num_undo_prompts: automatically undo bad moves when undo is on at most this many times. Can be a fraction like 0.5 for 50% chance of being granted an undo on a bad move.

FAQ

• The program is slow to start!
  • The first startup of KataGo can be slow, after that it should be much faster.
• The program is running too slowly!
  • Lower the visits count in the max_visits block of config.json by half or so and try again.

Contributing

• Feedback and pull requests are both very welcome.
• For suggestions and planned improvements, see the 'issues' tab on github.