def humanvmodel(sess, X, Y, humanfirst=False):
game = Game()
if humanfirst:
game.printState()
check = game.move(int(input("You go first...")))
while True:
if humanfirst:
pos = np.concatenate((game.noughts, game.crosses))
else:
pos = np.concatenate((game.crosses, game.noughts))
out = sess.run(Y, feed_dict={X: pos})
bestprob, bestmove = -10000.0, 0
for i in range(0, len(out)):
if out[i] > bestprob and game.spots[0, i] == 0.0:
bestprob = out[i]
bestmove = i
print(bestprob)
check = game.move(bestmove)
game.printState()
if check != 0.0:
print("Game over!")
break
check = game.move(int(input("Your turn...")))
if check != 0.0:
game.printState()
if check == 1000.0:
print("It's a draw!")
else:
print("You win!")
break
def train(sess, X, Y, optimizer, cost, boards, outcomes, ends, wons, moves,
explore_rate, memsize, batchsize, saver, directory, chckptrate,
Ytest):
boardm = np.zeros((memsize, 84))
outcomem = np.zeros((memsize, 84))
endm = np.zeros((memsize, 1))
wonm = np.zeros((memsize, 1))
movem = np.zeros((memsize, 1), dtype=int)
movem = movem - 1
test = np.zeros((memsize, 84))
game = Game()
firsttrain = True
counter = 0
noughtmoveip = False
boardtemp = np.zeros((1, 84))
movetemp = 0
boardtemp2 = np.zeros((1, 84))
movetemp2 = 0
avcost = 0
print("Beginning training...")
while True:
for i in range(0, chckptrate):
boardtemp = np.concatenate((game.crosses, game.noughts)).T
out = sess.run(Y, feed_dict={X: boardtemp.T})
movetemp = findbestmove(out, game, explore_rate)
check = game.move(movetemp)
#game.printState()
#input("Press any key")
if check == 1.0:
counter = addtomemory(boardm, outcomem,
endm, wonm, movem, boardtemp,
np.zeros((1, 84)), 1.0, 1.0, movetemp,
counter)
if noughtmoveip:
counter = addtomemory(boardm, outcomem, endm,
wonm, movem, boardtemp2,
np.zeros((1, 84)), 1.0, -1.0,
movetemp2, counter)
game.reset()
noughtmoveip = False
#print("Crosses won")
if check == 1000.0:
counter = addtomemory(boardm, outcomem,
endm, wonm, movem, boardtemp,
np.zeros((1, 84)), 1.0, 0.0, movetemp,
counter)
if noughtmoveip:
counter = addtomemory(boardm, outcomem, endm,
wonm, movem, boardtemp2,
np.zeros((1, 84)), 1.0, 0.0,
movetemp2, counter)
game.reset()
noughtmoveip = False
#print("Draw")
if check == 0.0:
if noughtmoveip:
counter = addtomemory(
boardm, outcomem, endm, wonm, movem, boardtemp2,
np.concatenate((game.noughts, game.crosses)).T, 0.0,
0.0, movetemp2, counter)
noughtmoveip = True
boardtemp2 = np.concatenate((game.noughts, game.crosses)).T
out = sess.run(Y, feed_dict={X: boardtemp2.T})
movetemp2 = findbestmove(out, game, explore_rate)
check2 = game.move(movetemp2)
#game.printState()
#input("Press any key")
if check2 == -1.0:
counter = addtomemory(boardm, outcomem, endm,
wonm, movem, boardtemp,
np.zeros((1, 84)), 1.0, -1.0,
movetemp, counter)
counter = addtomemory(boardm, outcomem, endm,
wonm, movem, boardtemp2,
np.zeros((1, 84)), 1.0, 1.0,
movetemp2, counter)
game.reset()
noughtmoveip = False
#print("Noughts won")
if check2 == 1000.0:
counter = addtomemory(boardm, outcomem, endm,
wonm, movem, boardtemp,
np.zeros((1, 84)), 1.0, 0.0,
movetemp, counter)
counter = addtomemory(boardm, outcomem, endm,
wonm, movem, boardtemp2,
np.zeros((1, 84)), 1.0, 0.0,
movetemp2, counter)
game.reset()
noughtmoveip = False
#print("Draw")
if check2 == 0.0:
counter = addtomemory(
boardm, outcomem, endm, wonm, movem, boardtemp,
np.concatenate((game.crosses, game.noughts)).T, 0.0,
0.0, movetemp, counter)
if movem[memsize - 1] != -1:
if firsttrain:
test = np.copy(boardm)
test = test[np.random.choice(np.arange(memsize),
size=1000,
replace=False)]
firsttrain = False
sample = np.random.choice(np.arange(memsize),
size=batchsize,
replace=False)
boardsample = boardm[sample]
outcomesample = outcomem[sample]
endsample = endm[sample]
wonsample = wonm[sample]
movesample = movem[sample]
_, batchcost = sess.run(
[optimizer, cost],
feed_dict={
boards: boardsample.T,
outcomes: outcomesample.T,
ends: endsample.T,
wons: wonsample.T,
moves: movesample.T
})
avcost = 0.9 * avcost + 0.1 * batchcost
if not firsttrain:
testvalue = np.average(
sess.run(Ytest, feed_dict={outcomes: test.T}))
print("Counter is " + str(counter) +
" and average Qmax on test set is " + str(testvalue))
saver.save(sess, directory)
# -*- coding: utf-8 -*-
"""
Created on Wed May 24 21:22:30 2017
@author: Toby
"""
from connectfour import Game
turncounter = 1
game = Game()
while True:
game.printState()
check = game.move(int(input("No winner yet...")))
if check != 0.0:
game.printState()
print("End of Game! Player " + str(check) + " has won!")
break
turncounter = turncounter + 1