def testing(template=None):
global bestcount
global best
if (template == None):
models1 = [evModel(9, 9, [16, 16, 9]) for k in range(200)]
models2 = [evModel(9, 9, [16, 16, 9]) for k in range(200)]
else:
models1 = [evModel(9, 9, [16, 16, 9], template[0]) for k in range(100)]
models2 = [evModel(9, 9, [16, 16, 9], template[1]) for k in range(100)]
models = zip(models1, models2)
bestmods1 = []
bestmods2 = []
for m1, m2 in models:
m1.mutate(3, 2)
m2.mutate(3, 2)
best = ticktactoe(m1, m2)
bestmods1.append([best[0], m1])
bestmods2.append([best[1], m2])
bestmods1 = sorted(bestmods1, key=itemgetter(0))
bestmods2 = sorted(bestmods2, key=itemgetter(0))
print bestmods1
return [
bestmods1[0][1].structure,
generateAverages([M[1] for M in bestmods2[len(bestmods2) / 4:]])
]
def ticktactoe(player1=None, player2=None):
global lastwinner
global streak
if (player1 == None):
player1 = evModel(9, 9, [16, 16, 9])
player2 = evModel(9, 9, [16, 16, 9])
player1.mutate(2, 2)
player2.mutate(2, 2)
game = [0] * 9
turns = 0
end = False
winner = player1
c = 0
while not end:
p1move = player1.run(game)
p1move = p1move.index(max(p1move))
if (game[p1move] != 0):
print "player failed"
return [c, 10 + c]
else:
print "move played"
game[p1move] = 1
if gameover(1, game):
if (lastwinner == 1):
streak += 1
else:
streak = 1
lastwinner = 1
print "WINNER 1"
return [1000 + (-100 * c), (c)]
p2move = player2.run(game)
p2move = p2move.index(max(p2move))
if (game[p2move] != 0):
print "player failed"
return [10 + c, c]
else:
print "move played"
game[p2move] = -1
if gameover(-1, game):
if (lastwinner == -1):
streak += 1
else:
streak = 1
lastwinner = -1
print "WINNER 2"
return [(c), 1000 + (-100 * c)]
if (sum(game) == 7):
print "TIE"
return [200, 200]
print game
c += 1
def testing(template=None):
global images
global labels
global bestcount
global best
if (template == None):
models = [evModel(378, 10, 5, activation=1) for k in range(25)]
for model in models:
model.mutate(2, 4)
else:
models = [
evModel(378, 10, 5, template[k], activation=1)
for k in range(len(template))
]
bestmods = []
outset = []
for model in models:
model.mutate(.5, 4)
c = 0
k = 0
seed = [random.randint(0, 125) for z in range(100)]
tests = [images[seed[z]] for z in range(8)]
answers = [labels[seed[z]] for z in range(8)]
outs = []
for test, answer in zip(tests, answers):
print c
out = model.run(test)
aindex, value = max(enumerate(out), key=operator.itemgetter(1))
print str(aindex) + "|" + str(answer)
outs.append(aindex)
k += (out[answer]) * 10
del out[answer]
k -= 1 - sum(out)
print k
c += 1
bestmods.append((k, model))
if k > bestcount:
bestcount = k
best = model
if (len(set(outset)) == 1):
print "genocide"
crazyMod = evModel(378, 10, 5, activation=1)
crazyMod.mutate(.05, 50)
return generateAverages([crazyMod])
bestmods = sorted(bestmods, key=itemgetter(0))
print bestmods
return generateAverages([M[1] for M in bestmods[len(bestmods) / 2:]])
def playticktactoe(player1=None, ):
if (player1 == None):
player1 = evModel(9, 9, [16, 16, 9])
player2 = evModel(9, 9, [16, 16, 9])
player1.mutate(5, 10)
player2.mutate(5, 10)
game = [0] * 9
turns = 0
end = False
winner = player1
c = 0
while not end:
p1move = player1.run(game)
p1move = p1move.index(max(p1move))
if (game[p1move] != 0):
print "player failed"
return [-100 + c, 10 + c]
else:
print "move played"
game[p1move] = 1
if gameover(1, game):
print "WINNER 1"
return [1000 + (-100 * c), -c * 2]
p2move = int(raw_input("play"))
if (game[p2move] != 0):
print "player failed"
return [10 + c, -100 + c]
else:
print "move played"
game[p2move] = -1
if gameover(-1, game):
print "WINNER 2"
return [c * 2, 1000 + (-100 * c)]
if (sum(game) == 7):
print "TIE"
return [200, 200]
print game
c += 1
from evolution import evModel from evolution import generateAverages model = evModel(3, 2, [2, 2]) print model.structure model.mutate(1, 2) print model.run([0, 1, 0]) model1 = evModel(3, 2, [2, 2]) model1.mutate(1, 2) print generateAverages([model, model1])
import numpy as np
import random
from sklearn.datasets import load_digits
from operator import itemgetter
from mnist import MNIST
from evolution import evModel
from evolution import generateAverages
import math
best = evModel(6, 3, 6, activation=1)
bestcount = 0
import operator
mndata = MNIST('MNIST')
images, labels = mndata.load_training()
newimages = []
for image in images[:200]:
temp = []
c = 0
while c < 729:
if (c > 27 and c % 27 == 0):
c += 27
if (c + 27 < 729):
temp.append(
max([
np.tanh(image[c] * 1.0),
np.tanh(image[c + 1] * 1.0),
np.tanh(image[c + 27] * 1.0),
np.tanh(image[c + 28] * 1.0)
]))
c += 1
if (game[p1move] != 0):
print "player failed"
return [-100 + c, 10 + c]
else:
print "move played"
game[p1move] = 1
if gameover(1, game):
print "WINNER 1"
return [1000 + (-100 * c), -c * 2]
p2move = int(raw_input("play"))
if (game[p2move] != 0):
print "player failed"
return [10 + c, -100 + c]
else:
print "move played"
game[p2move] = -1
if gameover(-1, game):
print "WINNER 2"
return [c * 2, 1000 + (-100 * c)]
if (sum(game) == 7):
print "TIE"
return [200, 200]
print game
c += 1
last = testing()
for k in range(1000):
last = testing(last)
playticktactoe(evModel(9, 9, [16, 16, 9], last[0]))