def test_getOutput(self):
nn = neuralNet.NeuralNet(4, 4, 1, 6)
numberOfWeights = (4 + 1) * 6 + (6 + 1) * 4
weights = [1 for i in range(numberOfWeights)]
nn.putWeights(weights)
in_ = [1, 1, 1, 1]
self.assertEqual(len(nn.getOutput(in_)), 4)
def main():
start = time.clock()
numInput, numOutput, numHiddenLayer, neuronPerLayer = 8, 1, 1, 2
net = NN.NeuralNet(numInput, numOutput, numHiddenLayer, neuronPerLayer)
# ga = GA.GenAlg()
with open('ga_tftrain_sig.pickle', 'rb') as fp:
print(fp.name)
ga = pickle.load(fp)
end = time.clock()
runtime = end - start
print("init net and GA: " + str(runtime))
width = 8 # of board
height = 8 # of board
games = 100
player1wins = 0
for i in range(0, games):
print("game num: " + str(i))
start = time.clock()
if play(player1, player2, width, height, ga, net) == 1:
player1wins += 1
end = time.clock()
runtime = end - start
print("runtime is " + str(runtime))
# pickle.dump(ga, open('ga-data_step.pickle', 'wb'))
return 'Player 1 won ' + str(player1wins) + ' out of ' + str(
games) + ' games of size ' + str(height) + 'x' + str(
width) + '. Total Game Time: ' + str(runtime) + ' sec'
def createNetwork(createFile, trainFromFile, nnParameters):
''' TRAIN DATA PROCESSING '''
train_data_file_name = 'data/spotifyTrainData.csv'
print "Train from file: ", trainFromFile, not trainFromFile
if not trainFromFile:
access_token = waitForAccessToken()
sa.crawlSpotifyData(access_token)
sa.labelSpotifyData(access_token)
if createFile:
createDataFileFromSpotify(train_data_file_name)
else:
if trainFromFile:
train_data, train_label = getTrainDataFromFile(
train_data_file_name)
else:
train_data, train_label = getTrainDataFromSpotify()
''' NEURAL NETWORK '''
nn = NN.NeuralNet(numInputs=nnParameters[0],
layers=nnParameters[1],
numOutputs=nnParameters[2],
learningRate=nnParameters[3],
momentum=nnParameters[4])
nn.printArchitecture()
nn.train(train_data[:800],
train_label[:800],
num_epoch=nnParameters[5])
nn.test(train_data[800:], train_label[800:])
def test_putWeights(self):
nn = neuralNet.NeuralNet(4, 4, 1, 6)
numberOfWeights = (4 + 1) * 6 + (6 + 1) * 4
weights = [1 for i in range(numberOfWeights)]
nn.putWeights(weights)
for neuron in nn.layers[0].neurons:
self.assertEqual(len(neuron.weights), 5)
for neuron in nn.layers[1].neurons:
self.assertEqual(len(neuron.weights), 7)
def train1(numOfTraining):
width = 8 # of board
height = 8 # of board
with open('true.pickle', 'rb') as fp:
trueboard = list(pickle.load(fp))
with open('false.pickle', 'rb') as fp:
falseboard = list(pickle.load(fp))
numInput, numOutput, numHiddenLayer, neuronPerLayer = 8, 1, 1, 8
net = NN.NeuralNet(numInput, numOutput, numHiddenLayer, neuronPerLayer)
ga = GA.GenAlg()
# with open('ga-data_step.pickle', 'rb') as fp:
# print(fp.name)
# ga = pickle.load(fp)
for j in range(numOfTraining):
start = time.clock()
print("iteration #: " + str(j))
board = initial_board(width, height)
winboard.detranslate(random.choice(falseboard), board)
for genome in ga:
net.putWeights(genome.weights)
move1 = net.move(board, True)
genome.fitness = ga.Fitness(0.0, move1[0])
for genome in ga:
print(genome.fitness, end=" ")
print()
ga.Epoch()
winboard.detranslate(random.choice(trueboard), board)
for genome in ga:
net.putWeights(genome.weights)
move1 = net.move(board, True)
genome.fitness = ga.Fitness(1.0, move1[0])
for genome in ga:
print(genome.fitness, end=" ")
print()
ga.Epoch()
end = time.clock()
runtime = end - start
print("runtime is " + str(runtime))
pickle.dump(ga, open('ga_tftrain_sig.pickle', 'wb'))
def __init__(self,population,numOfInputs,numOfOutputs,
numOfHiddenLayers,neuronsPerHiddenLayer,
crossoverRate,mutationRate,
geneInit = chromosome.geneFunction):
self.population = population
self.generation = 1
self.chromosomeLength = neuronsPerHiddenLayer*(numOfInputs+1)
self.chromosomeLength += (neuronsPerHiddenLayer+1)*numOfOutputs
self.chromosomeLength += sum([neuronsPerHiddenLayer*(neuronsPerHiddenLayer+1)
for i in xrange(1,numOfHiddenLayers)])
self.geneticAlg = geneAlg.GeneticAlgorithm(population,
crossoverRate,mutationRate,self.chromosomeLength,geneInit)
self.neuralNets = [neuralNet.NeuralNet(numOfInputs,numOfOutputs,
numOfHiddenLayers,neuronsPerHiddenLayer) for i in xrange(population)]
for nNet, chromo in zip(self.neuralNets,self.geneticAlg.chromosomes):
nNet.putWeights(chromo.genes)
def main():
with open('true.pickle', 'rb') as fp:
trueboard = list(pickle.load(fp))
with open('false.pickle', 'rb') as fp:
falseboard = list(pickle.load(fp))
print(len(trueboard), len(falseboard))
nn = neuralNet.NeuralNet(0, 0, 0, 0)
data = []
for board in trueboard:
temp_board = training.initial_board(8, 8)
winboard.detranslate(board, temp_board)
data.append(np.array(nn.translate(temp_board)))
for board in falseboard:
temp_board = training.initial_board(8, 8)
winboard.detranslate(board, temp_board)
data.append(np.array(nn.translate(temp_board)))
print(len(data))
data = np.array(data)
k = 2
km = Kmeans()
clusters = km.assignment(data, k, trueboard, falseboard, nn)
i = 0
while not clusters.sholdStop():
km.oneIter(data, clusters)
print("iter num " + str(i))
# if i%10 == 0:
# clusters.plot("C:/Users/Adiel/Desktop/figures/" + "fig_" + str(i) + ".png")
i += 1
pickle.dump(clusters, open("clusters.pickle", "wb"))
for cluster in clusters:
print(len(cluster))
import neuralNet as nn import tester as test # create a new neural net input_nodes = 3 output_nodes = 3 hidden_nodes = 3 learning_rate = 0.3 net = nn.NeuralNet(input_nodes, output_nodes, hidden_nodes, learning_rate)
def test_numberOfLayers(self):
nn = neuralNet.NeuralNet(4, 4, 1, 6)
self.assertEqual(len(nn.layers), 2)
nn = neuralNet.NeuralNet(4, 4, 3, 6)
self.assertEqual(len(nn.layers), 4)
def test_NeuralNetCreate(self):
nn = neuralNet.NeuralNet(4, 4, 1, 6)
def test_numberOfNeurons(self):
nn = neuralNet.NeuralNet(4, 4, 1, 6)
self.assertEqual(len(nn.layers[0].neurons), 6)
self.assertEqual(len(nn.layers[1].neurons), 4)
import numpy as np
import pickle
# from sklearn import svm
from sklearn import linear_model
import random
import neuralNet as NN
import training as trn
import winboard
with open('true.pickle', 'rb') as fp:
trueboard = list(pickle.load(fp))
with open('false.pickle', 'rb') as fp:
falseboard = list(pickle.load(fp))
nn = NN.NeuralNet(0, 0, 0, 0)
model = linear_model.Perceptron(penalty=None,
alpha=0.0001,
fit_intercept=True,
n_iter=5,
shuffle=True,
verbose=0,
eta0=1.0,
n_jobs=1,
random_state=0,
class_weight=None,
warm_start=False)
# with open('model.pickle', 'rb') as fp:
# model = pickle.load(fp)
boards = []
tag = []
width, height = 8, 8
i = 0
#%%
import numpy as np
import neuralNet as n
import matplotlib.pyplot as plt
X = np.array([[0,0,1,0,1,0,1],[0,1,1,0,0,0,0],[1,0,1,0,0,1,0],[1,1,1,1,1,1,1]])
y = np.array([[0],[1],[1],[0]])
nn = n.NeuralNet(7,1,8)
print(y)
nn.trainLoop(X,y,2000)
show=False
if(show):
plt.plot(nn.plotPointsX[10:],nn.plotPointsY1[10:],c='red')
plt.plot(nn.plotPointsX[10:],nn.plotPointsY2[10:],c='blue')
for ind,o in enumerate(nn.output):
print("the answer "+str(ind)+" is "+str(o[0]))
plt.show()
nn.answer([0,1,1,0,0,0,0])
def train3(numOfTraining):
width = 8 # of board
height = 8 # of board
with open('true.pickle', 'rb') as fp:
trueboard = list(pickle.load(fp))
with open('false.pickle', 'rb') as fp:
falseboard = list(pickle.load(fp))
# with open('net.pickle', 'rb') as fp:
# net = pickle.load(fp)
numInput, numOutput, numHiddenLayer, neuronPerLayer = 8, 1, 1, 48
net = NN.NeuralNet(numInput, numOutput, numHiddenLayer, neuronPerLayer)
for j in range(numOfTraining):
# start = time.clock()
print("iteration #: " + str(j))
trueboards = random.sample(trueboard, 400)
falseboards = random.sample(falseboard, 100)
for i in range(400):
tb = initial_board(width, height)
winboard.detranslate(trueboards[i], tb)
net.train(tb, [1.0])
if (i % 4):
fb = initial_board(width, height)
winboard.detranslate(falseboards[i % 100], fb)
net.train(fb, [0.0])
# end = time.clock()
# runtime = end - start
# print("runtime is " + str(runtime))
# trueboards = random.sample(trueboard, 100)
# falseboards = random.sample(falseboard, 100)
avg_true = 0
avg_false = 0
avg_ans_false = 0
avg_ans_true = 0
trueboards = random.sample(trueboard, 100)
falseboards = random.sample(falseboard, 100)
for i in range(100):
tb = initial_board(width, height)
fb = initial_board(width, height)
winboard.detranslate(trueboards[i], tb)
winboard.detranslate(falseboards[i], fb)
move1 = net.move(tb, True)
move2 = net.move(fb, True)
avg_ans_true += move1[0]
avg_ans_false += move2[0]
if math.fabs(move1[0] - 1.0) <= 0.1:
avg_true += 1
if math.fabs(move2[0] - 0.0) <= 0.1:
avg_false += 1
print("avg_true: " + str(avg_true / 100) + " avg_false: " +
str(avg_false / 100))
print("avg_true_ans: " + str(avg_ans_true / 100) + " avg_false_ans: " +
str(avg_ans_false / 100))
pickle.dump(net, open('net.pickle', 'wb'))
def train2(numOfTraining):
width = 8 # of board
height = 8 # of board
with open('true.pickle', 'rb') as fp:
trueboard = list(pickle.load(fp))
with open('false.pickle', 'rb') as fp:
falseboard = list(pickle.load(fp))
numInput, numOutput, numHiddenLayer, neuronPerLayer = 8, 1, 1, 2
net = NN.NeuralNet(numInput, numOutput, numHiddenLayer, neuronPerLayer)
ga = GA.GenAlg()
# elite = set()
# with open('elite.pickle', 'rb') as fp:
# print(fp.name)
# print(sorted(pickle.load(fp).population, key=GA.Genome.getFitness)[-1:-101])
# ga.population = sorted(pickle.load(fp).population, key = GA.Genome.getFitness)[-1:-101]
# random.shuffle(ga.population)
# for genome in ga.population:
# genome.fitness = 0.0
for j in range(numOfTraining):
start = time.clock()
print("iteration #: " + str(j))
trueboards = random.sample(trueboard, 100)
falseboards = random.sample(falseboard, 100)
for i in range(100):
tb = initial_board(width, height)
fb = initial_board(width, height)
winboard.detranslate(trueboards[i], tb)
winboard.detranslate(falseboards[i], fb)
for genome in ga:
net.putWeights(genome.weights)
move1 = net.move(tb, False)
move2 = net.move(fb, False)
if move1[0] == 1.0:
genome.fitness += 1
if move2[0] == 0.0:
genome.fitness += 1
ga.population.sort(key=GA.Genome.getFitness)
ga.population.reverse()
for genome in ga:
print(genome.fitness, end=" ")
if genome.fitness == 200:
print("solution is genome " + str(ga.population.index(genome)))
pickle.dump(ga, open('ga_data_tftrain_step.pickle', 'wb'))
# ga.population = list(elite)
# pickle.dump(ga, open('elite.pickle', 'wb'))
exit()
# elif genome.fitness > 100:
# elite.add(genome.copy())
# print()
# print(len(elite))
if j == numOfTraining - 1:
pickle.dump(ga, open('ga_data_tftrain_step.pickle', 'wb'))
# print(len(elite))
# ga.population = list(elite)
# pickle.dump(ga, open('elite.pickle', 'wb'))
exit()
ga.Epoch()
end = time.clock()
runtime = end - start
print("runtime is " + str(runtime))