If you have some requests, please let me know through my Twitter (@Nyanyan_Cube or @takuto_yamana), GitHub issue, or pull requests.
Functions below are in Study mode. Some functions are disabled in other modes.
Check if the all things are in these directories before you start the application.
Egaroucid5.exe
resources directory
book.egbk
eval.egev
joseki.txt
settings.txt
img directory
checked.png
icon.png
logo.png
languages directory
english.json
japanese.json
languages.json
languages.txt
records directory
Execute Egaroucid5.exe
to start the application.
Lookahead depth is defined by the "Levels". These levels are defined as this image.
You have to be careful when raising the level, because some high levels are too heavy to use in your computer.
Evaluation values are shown in each legal cells.
If the board is in your book, book
will be shown. If not, the AI will search at the Hint Level
and show the the level or 100%
when exact search is done. The best moves are shown in blue letters.
This application has another evaluation values. Human Sense Values are defined as putting together these values:
The idea came from (in Japanese): https://othelloq.com/tweet/quantifing-human-difficulty
CheckingHuman Sense Value
to enable this function. The values are shown in upper-right corner. The greater the value is, the better the move is.
Human Sense Values are calculated as:
Umigame's values are the numbers of the bifurcations of the best moves to memorize in your book.
You can see this value with checking Umigame's value
.
Black letters show black player's number, White letters show white player's number.
Umigame's values are announced here in Japanese: http://blog.livedoor.jp/umigame_oth/archives/1075469317.html
Clicking Analyze
button, the evaluation values are re-calculated and plotted on the graph in AI Level
.
You can move across the boards with dragging the graph or pushing right/left or A/D keys. You can play another line with putting a stone.
You can add a record in your clipboard. The record must be written in F5D6 or f5d6 format.
If importing succeed, you can see the board in your app.
You can add a board in your clipboard. The board must be written in one line with these characters.
0/B/b/X/x/*
1/W/w/O/o
./-
The color to play must be put in the 65th character.
The sample input is:
xxxxxxxxxx
...........................10......100.....1....................0
If importing succeed, you can see the board in your app.
You can save the game with clicking Save Game
button. You can add white/black player names and memo.
A text file will be created in records
directory.
The filename is:
xyyyy_mm_dd_hh_mm_ss.txt
yyyy: Year
mm: Month
dd: Day
hh: Hour
mm: Minute
ss: Second
The contents:
xxxxxxxxxx
f5d6 record
Black's Score
yyyy_mm_dd_hh_mm_ss
AI Level
Playing Mode(0:AI plays second 1:AI plays second 2:AI vs AI 3:Human vs Human)
Black player's name
White player's name
Memo
If the game is not finished, the score will be ?
.
You can copy the record of the board shown with clicking Copy Record
button.
If AI found at least one move of the legal moves, the AI do not search and put a stone from registered moves. That means that if you register bad moves only in your book, the AI will put on bad squares. You must register better moves first.
When some legal moves from a board are registered in your book with different evaluation values, AI puts a disc with book error.
Let X as the value of the best move in the legal positions and let Y as the book error, the policy is selected from legal moves in book which has (X-Y) or higher values.
If the registered moves are +1, -2, -10 respectively and book error is 0, to 2, The +1 move only selected. If book error is 3 to 10, one move is selected from +1 or -2 moves randomly. if book error is over 10, a move is selected from all moves.
Set any board and right click the legal cell. You can now set a new value with keys number keys, ten keys, and -
key. You see what you typed on lower left of the app.
If you right click a legal cell by mistake, do nothing and click the cell again. You can escape.
When closing the app, the book is automatically saved.
First, you have to set a board which will be the root of a book.
Then you have to set AI Level
now because this cannot be changed after starting learning the book.
Also You can set the depth of the book and disc difference acceptance now.
Then press the Start Learning
to start.
You can stop it anytime by Stop Learning
.
Books are created with the algorithm below.
The algorithm uses priority queue.
Any book created with this application or created by Edax can be imported on this app.
Press Import
button in the Book
tab, then you can import it with drag-and-drop.
Set any board as the root of the modification, then press Auto Modification
to modify the values after the board.
You can see that the latter value has better accuracy than the early one. This function uses latter values to modify the early values.
Best moves must be in your book. If not, the values after modification will be different from what you expected.
This application has a default book. This book is created as:
I got a permission to use Zebra's book.
Josekis are shown in Japanese. If you created a Joseki data, then I'm very pleased if you send me to pack it in the downloaded zip file.
I used the joseki data here: http://evaccaneer.livedoor.blog/archives/11101657.html
This Application is released under GPL3.0 license. you can use freely under this license.
Here you can see codes. Here is a code with which you can use AI without GUI. Please compile it to use it.
Inputting format is:
xxxxxxxxxx
Player number(0: Black 1: White)
Board Row 1
Board Row 2
Board Row 3
Board Row 4
Board Row 5
Board Row 6
Board Row 7
Board Row 8
For example,
xxxxxxxxxx
0
........
........
........
...10...
...100..
...1....
........
........
Outputting format is:
xxxxxxxxxx
{Row of the next move (0-7)} {Column of the next move (0-7)} {evaluation value}