Releases: kaorahi/lizgoban
LizGoban 0.9.0
Highlights:
- Upgrade KataGo to 1.16.0.
- Add visualization of AI's search tree. (
Tool > Plot MCTS tree) sample - Add rank estimation feature if a human model is available:
- rank estimation (9d, 3d, 1k, 6k, or 15k, by default. Check
Edit > Preferences > Finer dan/kyu scanfor finer but slower estimation.) - each rank's preferences in the winrate graph
- automatic adjustment of the human-style profile in match vs. AI.
- rank estimation (9d, 3d, 1k, 6k, or 15k, by default. Check
- [Experimental] Show the search tree for "if players try to capture/rescue this stone" by shift + double-click. You need to specify KataGo's option
-human-modelfor good results. (In the all-in-one package for Windows, choose "Human-like Analysis" from "Preset" menu.) ref
Further updates:
- Add board and stone images to the all-in-one package for Windows.
- Add board position copy-paste.
- Alt+drag to select the source region.
Edit > Flip / rotate / etc. > copy stones- Alt+drag to select the destination region.
Edit > Flip / rotate / etc. > paste
- Add tsumego frame without ko threats. (
Tool > Tsumego frame) - Add "Next move quiz" to View menu.
Incompatibilities with 0.8.*:
- Upgrade libraries (Electron 36, etc.). So you may need to do "npm install" again if you use LizGoban from the command line.
Human-style features
Choose "Human-like Analysis" or "Human-like Play" from "Preset" menu and refer to "KataGo" section in "Help" menu for details.
To use it on 64bit Windows immediately
Just download the all-in-one package (LizGoban-*_win_*.zip), unzip it, and double-click LizGoban *.exe. You do not need installation, configuration, additional downloads, and so on. Its file size is due to the built-in engine:
- KataGo (eigenavx2, opencl)
- 18 block network (kata1-b18c384nbt-s9996)
- human-trained network (b18c384nbt-humanv0)
You can switch KataGo versions (CPU and GPU) by [Preset] menu in LizGoban. The first run of the GPU version may take a long time (1 hour on a low-spec machine, for example) for its initial tuning. You can also choose "Human-like Analysis" or "Human-like Play" from [Preset] menu. Refer to "KataGo" section in [Help] menu for details.
To customize it on 64bit Windows
If you want to use another network (aka. model, weights), you can simply click the Engine menu and select "Load network weights". Additionally, you can modify the config.json file for more flexible configuration. See README for details.
To use it on other platforms (Mac, Linux, ...)
Download the source code and see README.md.
LizGoban 0.8.1-pre3
- Upgrade KataGo to 1.16.0.
- Add rank estimation in the winrate graph if a human model is available. (Check
Edit > Preferences > Finer dan/kyu scanfor finer but slower estimation.) - Add automatic adjustment of the rank in match vs. AI if a human model is available.
- Add board and stone images to the all-in-one package for Windows.
- Fix too strong HumanSL play in match vs. AI in 0.8.1-pre2.
(We have made substantial changes since 0.8.0, so we might call the next release 0.9.0 instead of 0.8.1.)
LizGoban 0.8.1-pre2
mcts_x4_large.mp4
(This video is at 4x real-time speed.)
- Add visualization of AI's search tree. (
Tool > Plot MCTS tree) sample - [Experimental] Show the search tree for "if players try to capture/rescue this stone" by shift + double-click. You need to specify KataGo's option
-human-modelfor good results. (In the all-in-one package for Windows, Choose "Human-like Analysis" from "Preset" menu.) ref - Add each rank's preferences in the winrate graph if a human model is available.
- Add board position copy-paste.
- Alt+drag to select the source region.
Edit > Flip / rotate / etc. > copy stones- Alt+drag to select the destination region.
Edit > Flip / rotate / etc. > paste
- Add tsumego frame without ko threats. (
Tool > Tsumego frame) - Upgrade libraries (Electron 35, etc.). So you may need to do "npm install" again if you use LizGoban from the command line.
LizGoban 0.8.1-pre1
Just a test of an experimental feature "next move quiz".
- Add "Next move quiz" to View menu.
- Upgrade libraries (Electron 33, etc.). So you may need to do "npm install" again if you use LizGoban from the command line.
LizGoban 0.8.0
Highlights:
- Upgrade KataGo to 1.15.3.
- Separate the model files from *.exe for larger models including a model for human-like style.
- Add "human-style" features.
- Compare the policies between 5kyu and 1dan, for example.
- Play human-like moves for the specified rank.
- Add a new strategy "spar" to "match vs. AI". It's focused on creating skill-testing situations.
- Add a sound feature.
- Add "Open recent" to File menu.
Further updates:
- Blur ownership display. (Borrow the idea from katrain#555.)
- Replace zone indicator with playing styles indicator.
- Add ownership distribution chart at the bottom left. (Press "x" key to enlarge it.)
- Add faint red rings around "hot" stones.
- Add thin red background for "hot" periods in winrate graph.
- Add "ambiguity of areas" (faint gray line) and "settled territories" (faint green/pink dots) to score graph.
- Warn overlooked high-policy best moves by squares on stones.
- Highlight settled areas by "v" key.
- Make long press of cursor keys smoother.
- Change playing style of persona strategy. This is still being tested and might change in the future.
- Avoid unnatural tenuki in match vs. weakened AI.
- Add random pair match.
- Detect encoding of SGF files etc.
- Support TamaGo.
- Deprecate the display of preferred moves by "AIs for handicap games".
- Deprecate homemade "aggressiveness" features and rely on the native KataGo features.
Incompatibilities from 0.7.*:
- Change the autosave format.
- Upgrade libraries (Electron 32, etc.). So you may need to do "npm install" again if you use LizGoban from the command line.
Human-style features
Choose "Human-like Analysis" or "Human-like Play" from "Preset" menu and refer to "KataGo" section in "Help" menu for details.
Thanks to dfannius; this analysis feature is a variation of his "policy heatmap".
You can also play against the "spar" AI. Designed for practice, it focuses not on winning but on creating skill-testing situations for players of specific DAN or KYU ranks. Let's knock it out!
- From "Preset" menu, choose "Human-like Play". ("Human-like Analysis" is also ok.)
- From "File" menu, choose "Match vs. AI".
- Select "spar" in "vs." pulldown menu.
- Adjust the profile slider. (20k-9d)
- Click the board to place the first black stone, or click "start AI's turn" button to let the AI play black.
To use it on 64bit Windows immediately
Just download the all-in-one package (LizGoban-*_win_*.zip), unzip it, and double-click LizGoban *.exe. You do not need installation, configuration, additional downloads, and so on. Its file size is due to the built-in engine:
- KataGo 1.15.3 (eigenavx2, opencl)
- 18 block network (kata1-b18c384nbt-s9996)
- human-trained network (b18c384nbt-humanv0)
You can switch KataGo versions (CPU and GPU) by [Preset] menu in LizGoban. The first run of the GPU version may take a long time (1 hour on a low-spec machine, for example) for its initial tuning. You can also choose "Human-like Analysis" or "Human-like Play" from [Preset] menu. Refer to "KataGo" section in [Help] menu for details.
To customize it on 64bit Windows
If you want to use another network (aka. model, weights), you can simply click the Engine menu and select "Load network weights". Additionally, you can modify the config.json file for more flexible configuration. See README for details.
To use it on other platforms (Mac, Linux, ...)
Download the source code and see README.md.
LizGoban 0.8.0-pre8
- Added stone sounds. You can disable them from the menu
View > SoundorEdit > Preferences. - Minor fixes, etc.
Incompatibilities:
- Upgraded libraries (Electron 32, etc.). So you may need to do "npm install" again if you use LizGoban from the command line.
LizGoban 0.8.0-pre7
An experimental strategy called "spar" has been added to the "match vs. AI" mode. Intended for practice, it's not focused on winning but on creating skill-testing situations for players of specific DAN or KYU ranks. Let's knock it out!
To test it:
- Download the all-in-one package (
LizGoban-*_win_*.zip), unzip it, and double-clickLizGoban *.exe. - From "Preset" menu, choose "Human-like Play". ("Human-like Analysis" is also ok.)
- From "File" menu, choose "Match vs. AI".
- Select "spar" in "vs." pulldown menu.
- Adjust the profile slider. (20k-9d)
- Click the board to place the first black stone, or click "start AI's turn" button to let the AI play black.
LizGoban 0.8.0-pre6
This is just a temporary release for a debug feature on the asymmetric rank humanSL profile. This feature will be excluded from the next release.
KataGo is upgraded to 1.15.3.
To test the asymmetric rank humanSL profile on Windows
(Sorry for the long steps. This is just a debug feature.)
- Download the all-in-one package (
LizGoban-*_win_*.zip), unzip it, and double-clickLizGoban *.exe. - From "Preset" menu, choose "Human-like Analysis".
- From "Tool" menu, choose "Experimental" > "Enable debug menu".
- From "Debug" menu, choose "Asymmetric humanSL rank" to toggle it on.
- From "Edit" menu, choose "Preferences" to open the dialog and adjust the sliders as you like. Toggle "Lizzie style" off if the display is hard to see.
For example, you can set the blue slider to rank_1d and the red slider to rank_5k to compare the policies between "1d" and "1d against 5k". The former indicates "I'm 1d", while the latter indicates "I'm 1d and I know you're 5k". These policies, p and q, are shown by blue or red squares on the board (size = max(p,q), color = log(p/q)). Blue squares mean p > q, and red squares mean p < q.
LizGoban 0.8.0-pre5
- Upgrade KataGo to 1.15.1.
- Separate the model files from *.exe for larger models including a model for human-like style.
- Add "human-style" features.
- Compare the policies between 5kyu and 1dan, for example.
- Play human-like moves for the specified rank.
- Warn overlooked high-policy best moves by squares on stones.
- Avoid too stupid moves in persona strategy.
- Add random pair match.
- Deleted boards are also listed in "Open recent" menu.
- Deprecate the display of preferred moves by "AIs for handicap games".
- Deprecate homemade "aggressiveness" features and rely on the native KataGo features.
Incompatibilities:
- Change the autosave format.
- Upgrade libraries (Electron 31, etc.). So you may need to do "npm install" again if you use LizGoban from the command line.
Human-style features
Choose "Human-like Analysis" or "Human-like Play" from "Preset" menu and refer to "KataGo" section in "Help" menu for details.
Thanks to dfannius; this analysis feature is a variation of his "policy heatmap".
To use it on 64bit Windows immediately
Just download the all-in-one package (LizGoban-*_win_*.zip), unzip it, and double-click LizGoban *.exe. You do not need installation, configuration, additional downloads, and so on. Its file size is due to the built-in engine:
- KataGo 1.15.1 (eigenavx2, opencl) + 18 block network (kata1-b18c384nbt-s9996) + human-trained network (b18c384nbt-humanv0.bin.gz)
You can switch KataGo versions (CPU and GPU) by [Preset] menu in LizGoban. The first run of the GPU version may take a long time (1 hour on a low-spec machine, for example) for its initial tuning. You can also choose "Human-like Analysis" or "Human-like Play" from [Preset] menu. Refer to "KataGo" section in [Help] menu for details.
To customize it on 64bit Windows
If you want to use another network (aka. model, weights), you can simply click the Engine menu and select "Load network weights". Additionally, you can modify the config.json file for more flexible configuration. See README for details.
To use it on other platforms (Mac, Linux, ...)
Download the source code and see README.md.
LizGoban 0.8.0-pre4
- Upgrade KataGo to 1.14.0.
- Add sound feature. (You need to prepare and set sound files by yourself.)
- Replace SOPPO indicator with playing styles indicator.
- Add playing styles indicator to ownership distribution chart.
- Change playing style of persona strategy. This is still being tested and might change in the future.
- Avoid unnatural tenuki in match vs. weakened AI.
- Add "Open recent" to File menu.
- Detect encoding of SGF files etc.
- Support TamaGo. (You need a patch currently on 2023-12-28.)
Minor changes:
- Add "board_size" and "wait_for_startup" to preset in config.json.
- Make
Ctrl-,key toggle preferences window. - Enable app.quit() in REPL for external control.
- Add "default preset" to "Engine is down" dialog for SSH disconnection.
- Use entropy as the ambiguity measure.
- Fix bugs.
Incompatibilities:
- Upgrade libraries (Electron 28, etc.). So you may need to do "npm install" again if you use LizGoban from the command line.
Acknowledgement:
- Thanks to qcgm1978 for the contributions and suggestions.