Skip to content

@lemonsqueeze lemonsqueeze released this Mar 31, 2019 · 2 commits to master since this release

Binary release for Linux / Windows.

Changes since 12.30:

  • Speedup, about 20% faster
  • Multiple dcnns support: detlef44, detlef54, darkforest
  • lz-genmove_analyze support (get winrates in sabaki etc)
  • lz-analyze: handle frequency
  • Set engine options over gtp
  • Dcnn pondering fix

Use pachi_19x19.exe if you only play on 19x19, should be slightly faster.

Edit: Apr 06 - Fixed darkforest, detlef44 dcnn issues, should work now.

Assets 5

@lemonsqueeze lemonsqueeze released this Feb 23, 2019 · 3503 commits to master since this release

Extra dcnns supported by Pachi:

  • Detlef's 54% dcnn
    Default network, about 1d level.
    Fully convolutional, so can work with boards other than 19x19.

  • Detlef's 44% dcnn
    Its predecessor.

  • Darkforest dcnn
    Trained to predict next 3 moves, about 3d level.
    Converted to caffe with torch2caffe.

Assets 5

@lemonsqueeze lemonsqueeze released this Feb 11, 2019 · 86 commits to master since this release

Binary release for Linux / Windows.

Changes since 12.20:

  • Lizzie support !
  • Dcnn pondering
  • Japanese rules: territory scoring in playouts
  • Full gtp undo

Use pachi_19x19.exe if you only play on 19x19, it should be slightly faster.

Assets 5

@lemonsqueeze lemonsqueeze released this Nov 4, 2018 · 139 commits to master since this release

Changes since 12.10:

  • dcnn: handle smaller board sizes
  • don't play slack when taking handicap
  • mm patterns fixes, better prediction rate
  • use joseki engine when playing without dcnn

If you only play on 19x19 use the binaries in extras.zip they should be slightly faster.

Edit: Fixed assertion issue

Assets 5

@lemonsqueeze lemonsqueeze released this Aug 5, 2018 · 187 commits to master since this release

Changes since 12.00:

  • mm patterns: better startup times, memory usage, and middle game reading.
  • playouts: handle bent-four, special sekis
  • infrastructure fixes

New pattern code makes it possible to play with few playouts without dcnn:
For example pachi --nodcnn -t =5000 is kgs 3k.

If you only play on 19x19 use the binaries in extras.zip they should be slightly faster.

Assets 4

@lemonsqueeze lemonsqueeze released this Mar 6, 2018 · 253 commits to master since this release

Changes since 11.00 release:

  • dcnn support
  • 32/64 bits support
  • improved multithreaded performance
  • handle huge max_tree_size on win64
  • new defaults

If you only play on 19x19 use the binaries in extras.zip they should be slightly faster.

Assets 4
Sep 8, 2017
board: BOARD_TRAITS
Sep 8, 2017
board: BOARD_SPATHASH

@lemonsqueeze lemonsqueeze released this Apr 7, 2019 · 496 commits to master since this release

Windows binary of stable Pachi version 11.00 kindly contributed by Neil McLean.

Download and extract large patterns in Pachi directory to make it 1-2 stones stronger.
Best is gogod-handikgspachi.

Assets 3

@lemonsqueeze lemonsqueeze released this Apr 7, 2019 · 570 commits to master since this release

Windows binary of stable Pachi version 10.00 kindly contributed by Neil McLean.

Download and extract large patterns in Pachi directory to make it 1-2 stones stronger.
See README_PATTERNS for details.
Best is gogod-handikgspachi.

Assets 7
You can’t perform that action at this time.