Changeset - 3cce3db6d4cc
[Not reviewed]
default
0 2 0
Laman - 7 years ago 2017-12-03 11:45:58

work on StateBag
2 files changed with 80 insertions and 1 deletions:
0 comments (0 inline, 0 general)
src/go.py
Show inline comments
 
import logging as log
 

	
 
from util import EMPTY,BLACK,WHITE,colorNames
 
from gamerecord import GameRecord
 

	
 

	
 
class Go:
 
	## Initializes self.board to a list[r][c]=EMPTY.
 
	def __init__(self,boardSize=19):
 
		self.boardSize=boardSize
 
		self.board=[[EMPTY]*boardSize for x in range(boardSize)]
 
		self._temp=[[EMPTY]*boardSize for x in range(boardSize)]
 
		self.toMove=BLACK
 
		self._record=GameRecord()
 

	
 
	## Executes a move.
 
	#
 
	#  Doesn't check for kos. Suicide not allowed.
 
	#
 
	#  @param color BLACK or WHITE
 
	#  @return True on success, False on failure (illegal move)
 
	def move(self,color,row,col):
 
		if color!=self.toMove: log.warning("move by %s out of order",colorNames[color])
 
		if self.board[row][col]!=EMPTY: return False
 

	
 
		self.board[row][col]=color
 

	
 
		# capture neighbors
 
		for r,c in ((-1,0),(1,0),(0,-1),(0,1)):
 
			self._clearTemp()
 
			if not self._floodFill(-color,row+r,col+c): self._remove()
 

	
 
		# check for suicide
 
		self._clearTemp()
 
		if not self._floodFill(color,row,col):
 
			self.board[row][col]=EMPTY
 
			return False
 
		self._record.move(color,row,col)
 
		self.toMove=-1*color
 
		return True
 

	
 
	def transitionMove(self,board):
 
		res=transitionMove(self.board,board)
 
		if not res: return res
 
		(r,c,color)=res
 
		return self.move(color,r,c)
 

	
 

	
 
	## Checks for liberties of a stone at given coordinates.
 
	#
 
	#  The stone's group is marked with True in self.temp, ready for capture if needed. Recursively called for stone's neighbors.
 
	#
 
	#  @return True if alive, False if captured
 
	def _floodFill(self,color,row,col):
 
		if col<0 or col>=self.boardSize or row<0 or row>=self.boardSize: return False # out of range
 
		if self._temp[row][col]: return False # already visited
 
		if self.board[row][col]==EMPTY: return True # found a liberty
 
		if self.board[row][col]!=color: return False # opponent's stone
 
		self._temp[row][col]=True # set visited
 
		return self._floodFill(color,row,col-1) or self._floodFill(color,row,col+1) or self._floodFill(color,row-1,col) or self._floodFill(color,row+1,col) # check neighbors
 

	
 
	## Removes stones at coordinates marked with True in self.temp.
 
	def _remove(self):
 
		for r in range(self.boardSize):
 
			for c in range(self.boardSize):
 
				if self._temp[r][c]: self.board[r][c]=EMPTY
 

	
 
	def _clearTemp(self):
 
		for i in range(self.boardSize):
 
			for j in range(self.boardSize):
 
				self._temp[i][j]=EMPTY
 

	
 

	
 
def exportBoard(board):
 
	substitutions={EMPTY:".", BLACK:"X", WHITE:"O"}
 
	return "\n".join("".join(substitutions.get(x,"?") for x in row) for row in board)
 

	
 

	
 
def isLegalPosition(board):
 
	boardSize=len(board)
 
	temp=[[None]*boardSize for x in range(boardSize)]
 

	
 
	for r in range(boardSize):
 
		for c in range(boardSize):
 
			if board[r][c]==EMPTY: continue
 
			if temp[r][c]: continue
 
			if not dfs([(r,c)],board,temp): return False
 
	return True
 

	
 

	
 
def transitionMove(state1,state2):
 
	moves=[]
 
	for (r,(row1,row2)) in enumerate(zip(state1,state2)):
 
		for (c,(item1,item2)) in enumerate(zip(row1,row2)):
 
			if item1==EMPTY and item2!=EMPTY:
 
				moves.append((r,c,item2))
 

	
 
	if len(moves)==0:
 
		log.info("no new stone")
 
		return None
 
	elif len(moves)==1:
 
		log.info("new stone: %s",moves[0])
 
		return moves[0]
 
	else:
 
		log.warning("too many new stones: %s",moves)
 
		return False
 

	
 

	
 
def transitionSequence(state1,state2,diff):
 
	return []
 

	
 

	
 
def dfs(stack,board,mask):
 
	boardSize=len(board)
 
	(r,c)=stack[0]
 
	color=board[r][c]
 

	
 
	while len(stack)>0:
 
		(r,c)=stack.pop()
 
		if board[r][c]==EMPTY: return True
 
		elif board[r][c]!=color: continue
 
		elif mask[r][c]: return True
 
		mask[r][c]=True
 
		for (x,y) in ((0,-1),(-1,0),(0,1),(1,0)):
 
			if 0<=r+y<boardSize and 0<=c+x<boardSize:
 
				stack.append((r+y,c+x))
 
	return False
src/statebag.py
Show inline comments
 
"""Theory:
 
We have a sequence S of valid board states s_1, ..., s_n.
 
We search for a picked subsequence S_p of length k and m additional moves M such that:
 
- S_p augmented by M forms a valid sequence of moves
 
- k-m is maximal for S_p picked from S
 
It is relatively cheap to add new items to S.
 

	
 
User can change detector parameters, presumably in case the previous don't fit the (current) situation.
 
In such a case we assume the new parameters are correct from the current position onwards.
 
But the change might have been appropriate even earlier (before the user detected and fixed an error).
 
So we try to find the correct crossover point like this:
 
- construct a sequence S' = s'_i, ..., s'_n by reanalyzing the positions with a new set of parameters, where s_i is the point of previous user intervention or s_0
 
- for each s'_j:
 
	- try to append it to S[:j]
 
	- try to append it to S'[:j]
 
	- remember the better variant
 
- linearize the fork back by discarding s'_j-s preceding the crossover and s_j-s following the crossover
 
"""
 

	
 
from util import BLACK,WHITE,EMPTY
 
from go import transitionSequence
 

	
 

	
 
## Crude lower bound on edit distance between states.
 
def estimateDistance(diff):
 
	additions=sum(1 for d in diff if d[2]=="+")
 
	deletions=sum(1 for d in diff if d[2]=="-")
 
	replacements=len(diff)-additions-deletions
 
	if additions>0: return additions+replacements
 
	elif replacements==0 and deletions>0: return 2 # take n, return 1
 
	return replacements # ???
 

	
 

	
 
class BoardState:
 
	def __init__(self,board):
 
		self._board=tuple(tuple(x for x in row) for row in board)
 
		self.prev=None
 
		self._next=None
 
		self.moves=[]
 
		self.weight=0
 
		self.diff2Prev=None
 

	
 
	def __iter__(self): return iter(self._board)
 

	
 
	def __getitem__(self,key): return self._board[key]
 

	
 
	def __sub__(self,x):
 
		res=[]
 

	
 
		for (r,(row1,row2)) in enumerate(zip(self._board,x)):
 
			for (c,(item1,item2)) in enumerate(zip(row1,row2)):
 
				if item1==item2: continue
 
				elif item2==EMPTY: res.append((r,c,"+",item1))
 
				elif item1==EMPTY: res.append((r,c,"-",item2))
 
				else: res.append((r,c,item2,item1)) # from->to
 
		return res
 

	
 

	
 
class StateBag:
 
	def __init__(self):
 
		self._states=[]
 

	
 
	def pushState(self,board):
 
		self._states.append(board)
 
		sn=BoardState(board)
 
		for s in reversed(self._states):
 
			diff=sn-s
 
			distEst=estimateDistance(diff)
 
			if distEst>3: continue # we couldn't find every such move sequence anyway without a clever algorithm
 
			weightEst=s.weight-distEst
 
			if weightEst<=sn.weight: continue
 
			moves=transitionSequence(s,sn,diff)
 
			weight=s.weight-len(moves)
 
			if weight<=sn.weight: continue
 
			sn.prev=s
 
			sn.diff2Prev=diff
 
			sn.moves=moves
 
			sn.weight=weight
 
		self._states.append(sn)
 

	
 
	def merge(self,branch):
 
		pass
0 comments (0 inline, 0 general)