class AI:
def __init__(self, inputFile=False, trainMode=False, realtimeMode=False, loadNeuralNetwork=False):
# Object variables for initialisation
self.inputFile = inputFile
self.trainMode = trainMode
self.realtimeMode = realtimeMode
self.loadNeuralNetwork = loadNeuralNetwork
self.neuralNet = FeedForwardNetwork()
inLayer = LinearLayer(7)
hiddenLayer = SigmoidLayer(5)
outLayer = LinearLayer(1)
self.neuralNet.addInputModule(inLayer)
self.neuralNet.addModule(hiddenLayer)
self.neuralNet.addOutputModule(outLayer)
self.neuralNet.addConnection(FullConnection(inLayer, hiddenLayer))
self.neuralNet.addConnection(FullConnection(hiddenLayer, outLayer))
self.neuralNet.sortModules()
DEBUG("Neural Network:")
DEBUG(self.neuralNet)
DEBUG("--------")
if self.loadNeuralNetwork:
self.neuralNet = NetworkReader.readFrom("lastNetwork.xml")
if self.inputFile:
if not os.path.exists(inputFile):
print "File for CSV parsing does not exist!"
return
savedData = self.readCSVData(self.inputFile)
# train upon the given data
if self.trainMode:
ds = self.makeTrainingSet(savedData)
for inpt, target in ds:
DEBUG("Input Values from training:")
DEBUG(inpt)
DEBUG("Correction factor from training:")
DEBUG(target)
trainer = BackpropTrainer(self.neuralNet, ds, verbose=True)
trainer.trainUntilConvergence(maxEpochs=10)
self.printGraph(ds)
NetworkWriter.writeToFile(self.neuralNet, "lastNetwork.xml")
# read the saved data and use it in the network
for line in savedData:
DEBUG("Used Data: ")
DEBUG(line)
result = self.neuralNet.activate(line)
DEBUG(result)
DEBUG("Final weigths:")
DEBUG(self.neuralNet.params)
# now fun begins. this is the standard mode this module runs in.
if self.realtimeMode:
print "We should start now!"
self.brain = Brain()
if os.path.isfile("lastNetwork.xml"):
while True:
lines = self.brain.getBins()
print (lines)
else:
# we do a realtime training of the neural network
print "No trained network exists!"
time.sleep(1)
print "Will do on-line training!"
time.sleep(1)
print "Hold tight for 10 seconds"
tempBinWriter = csv.writer(open("temp.csv", "ab"))
for i in range(1, 10):
time.sleep(1)
bins = self.brain.getBins()
print bins
tempBinWriter.writerow(bins)
tempSavedData = self.readCSVData("temp.csv")
ds = self.makeTrainingSet(tempSavedData)
trainer = BackpropTrainer(self.neuralNet, ds, verbose=True)
trainer.trainUntilConvergence(maxEpochs=10)
def getRandomness(self):
if self.realtimeMode:
bins = self.brain.getBins()
rand = 1
for i in range(7):
rand *= bins[i] * 100
return rand
def getNeuralResult(self):
return self.neuralNet.activate(self.brain.getBins())
def makeTrainingSet(self, savedData):
# init a dataset with seven input and one output value
ds = SupervisedDataSet(7, 1)
meanValue = self.getMeanValue(savedData)
correctionFactor = 0
for line in savedData:
# intelligent changing value for training a neural network
correctionFactor += (line[0] - meanValue[0]) * 3
correctionFactor += (line[1] - meanValue[1]) * 2
correctionFactor += line[2] - meanValue[2]
correctionFactor += meanValue[3] - line[3]
correctionFactor += (meanValue[4] - line[4]) * 2
correctionFactor += (meanValue[5] - line[5]) * 3
correctionFactor += (meanValue[6] - line[6]) * 6
DEBUG("Correction Factor:")
DEBUG(correctionFactor)
# copy values of current line for comparision
ds.addSample(line, correctionFactor)
correctionFactor = 0
return ds
def getMeanValue(self, savedData):
meanValues = [0] * 7
lines = 0
for line in savedData:
lines += 1
valuePosition = 0
for value in line:
meanValues[valuePosition] += value
valuePosition += 1
for i in range(7):
meanValues[i] /= lines
return meanValues
def printGraph(self, dataSet):
i = numpy.arange(0, len(dataSet["target"]), 1)
wavePercentage = dataSet["input"][i]
correctionFactor = dataSet["target"][i]
ax = plt.subplot(111)
plt.plot(i, wavePercentage)
plt.plot(i, correctionFactor + 0.15, c=cm.summer(0), linewidth=2)
xlabel(r"Number of Data Samples", fontsize=20)
box = ax.get_position()
ax.set_position([box.x0, box.y0, box.width * 0.8, box.height])
ax.legend(
("$Delta$ ", "$Theta$", "$Alpha$", "$Beta_1$", "$Beta_2$", "$Beta_3$", "$Gamma$", "$Target$"),
loc="center left",
bbox_to_anchor=(1, 0.5),
fancybox=True,
shadow=True,
)
show()
def readCSVData(self, inputFile):
lines = sum(1 for line in open(inputFile))
with open(inputFile, "rb") as f:
reader = csv.reader(f)
valuesRead = 0
# init the multidimensional array
values = numpy.zeros((lines, 7))
j = 0
for row in reader:
i = 0
for cell in row:
if i == 6:
values[j][i] = Decimal(cell)
valuesRead += 1
i = 0
else:
values[j][i] = Decimal(cell)
i += 1
j += 1
return values
