nemo
/
boardgame2vec
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Splits mcts code into separate file

This commit is contained in:
Nemo 2018-05-28 19:42:02 +05:30
parent 5449cc2a25
commit e38012b8fd
3 changed files with 109 additions and 100 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
gothok/.gitignore vendored
View File

@ -6,3 +6,4 @@ python/
*.pyx
*.so
*.bin
*.pyc

115
gothok/game.py
View File

@ -2,6 +2,7 @@
import capnp
import state_capnp as game
from mcts import GameNode
# https://blog.theofekfoundation.org/artificial-intelligence/2016/06/27/what-is-the-monte-carlo-tree-search/
@ -24,8 +25,9 @@ class State():
def playMove(self, move):
(xx, yy) = move
picked_card = self.board[xx][yy]
varys = self.findVarys
varys = self.findVarys()
vx = varys.board.x
vy = varys.board.y
@ -42,13 +44,20 @@ class State():
else:
raise Exception("Invalid move")
picked_cards = []
# TODO: switch this to a filter on cards_attempted
for (row, col) in cards_attempted:
card = self.board[row][col]
print(card.house, picked_card.house)
# If it is of the same house as declared
if (card == picked_card):
if (card.house == picked_card.house):
# Pick it up
state['cards'][current].append(picked_card)
state['board'][row][col] = self.EMPTY
picked_cards.append(card)
self.board[row][col] = game.House.empty
print(picked_cards)
def findVarys(self):
for card in self.state.cardlist:
@ -106,102 +115,8 @@ class State():
seen[str(house)] = True
class GameNode(object):
"""docstring for GameNode"""
def __init__(self, state, parent):
super(GameNode, self).__init__()
self.state = state
self.parent = parent
self.hits = 0
self.misses = 0
self.totalTrials = 0
def backPropagate(self, simulation):
if (simulation > 0):
self.hits += 1
elif (simulation < 0):
self.misses += 1
self.totalTrials += 1
if self.parent:
self.parent.backPropagate(-simulation)
def childPotential(self, child):
w = child.misses
n = child.totalTrials
# Chosen empirically
c = math.sqrt(2)
t = self.totalTrials
return (w / n) + (c * math.sqrt(log(t) / n))
def runSimulation(self):
self.backPropagate(self.simulate())
def simulate(self):
state = self.state
while not state.gameOver:
moves = state.getPossibleMoves()
randomMove = random.choice(possibleMoves)
state = state.playMove(randomMove)
return self.state.result(state)
def getChildren(self):
possibleMoves = self.state.getPossibleMoves()
children = []
for move in possibleMoves:
newState = self.state.playMove(move)
childNode = GameNode(newState, self.state)
children.append(childNode)
return children
def chooseChild(self):
# Define children nodes
if(not self.children):
self.children = self.getChildren()
# Run simulation on leaf nodes
if(len(self.children) == 0):
self.runSimulation()
else:
unexplored = []
# Get all unexplored nodes
for child in self.children:
if (child.totalTrials == 0):
unexplored.append(child)
# Pick a random unexplored node
# and run the simulation on it
if (len(unexplored) > 0):
random.choice(unexplored).runSimulation()
else:
# Find the best child
bestChild = self.children[0]
bestPotential = self.childPotential(bestChild)
for child in self.children:
potential = self.childPotential(child)
if (potential > bestPotential):
bestPotential = potential
bestChild = child
bestChild.chooseChild()
f = open('state.bin', 'rb')
initial_state = game.State.read_packed(f)
# print(initial_state)
s = State(initial_state)
print((s.getPossibleMoves()))
# 3,2
root_node = GameNode(s, None)
print(root_node.chooseChild())

93
gothok/mcts.py Normal file
View File

@ -0,0 +1,93 @@
import random
class GameNode(object):
"""docstring for GameNode"""
def __init__(self, state, parent):
super(GameNode, self).__init__()
self.state = state
self.parent = parent
self.hits = 0
self.misses = 0
self.totalTrials = 0
def backPropagate(self, simulation):
if (simulation > 0):
self.hits += 1
elif (simulation < 0):
self.misses += 1
self.totalTrials += 1
if self.parent:
self.parent.backPropagate(-simulation)
def childPotential(self, child):
w = child.misses
n = child.totalTrials
# Chosen empirically
c = math.sqrt(2)
t = self.totalTrials
return (w / n) + (c * math.sqrt(log(t) / n))
def runSimulation(self):
self.backPropagate(self.simulate())
def simulate(self):
state = self.state
while not state.gameOver:
moves = state.getPossibleMoves()
randomMove = random.choice(possibleMoves)
state = state.playMove(randomMove)
return self.state.result(state)
def getChildren(self):
possibleMoves = self.state.getPossibleMoves()
children = []
for move in possibleMoves:
newState = self.state.playMove(move)
childNode = GameNode(newState, self.state)
children.append(childNode)
return children
def chooseChild(self):
# Define children nodes
try:
self.children
except Exception as e:
self.children = self.getChildren()
if(len(self.children) == 0):
self.runSimulation()
else:
unexplored = []
# Get all unexplored nodes
for child in self.children:
if (child.totalTrials == 0):
unexplored.append(child)
# Pick a random unexplored node
# and run the simulation on it
if (len(unexplored) > 0):
random.choice(unexplored).runSimulation()
else:
# Find the best child
bestChild = self.children[0]
bestPotential = self.childPotential(bestChild)
for child in self.children:
potential = self.childPotential(child)
if (potential > bestPotential):
bestPotential = potential
bestChild = child
bestChild.chooseChild()