def __init__(self, nIn, nOut, weights=None,
activation='softmax', isClassifierLayer=True):
# Get activation function from string
# Notice the functional programming paradigms of Python + Numpy
self.activationString = activation
self.activation = Activation.getActivation(self.activationString)
self.activationPrime = Activation.getDerivative(self.activationString)
self.nIn = nIn
self.nOut = nOut
# Adding bias
self.input = np.ndarray((nIn+1, 1))
self.input[0] = 1
self.output = np.ndarray((nOut, 1))
self.delta = np.zeros((nOut, 1))
# You can have better initialization here
# wij means the weight from Input(j) to the Output(i)
if weights is None:
rns = np.random.RandomState(int(time.time()))
self.weights = rns.uniform(size=(nOut, nIn + 1))-0.5
else:
self.weights = weights
self.isClassifierLayer = isClassifierLayer
# Some handy properties of the layers
self.size = self.nOut
self.shape = self.weights.shape
def __init__(self,
nIn,
nOut,
weights=None,
activation='sigmoid',
isClassifierLayer=False):
# Get activation function from string
self.activationString = activation
self.activation = Activation.getActivation(self.activationString)
self.activationDerivative = Activation.getDerivative(
self.activationString)
self.nIn = nIn
self.nOut = nOut
self.inp = np.ndarray((nIn + 1, 1))
# self.inp[0] = 1
self.outp = np.ndarray((nOut, 1))
self.deltas = np.zeros((nOut, 1))
# You can have better initialization here
if weights is None:
rns = np.random.RandomState(int(time.time()))
self.weights = rns.uniform(size=(nIn + 1, nOut)) - 0.5
else:
assert (weights.shape == (nIn + 1, nOut))
self.weights = weights
self.isClassifierLayer = isClassifierLayer
# Some handy properties of the layers
self.size = self.nOut
self.shape = self.weights.shape
def __init__(self, train, valid, test,
learningRate=0.01, epochs=50,
activation='sigmoid',
error='mse'):
self.learningRate = learningRate
self.epochs = epochs
self.trainingSet = train
self.validationSet = valid
self.testSet = test
# Initialize the weight vector with small random values
# between -0.3 and 0.3 to encourage sigmoid function learning
self.weight = np.random.rand(self.trainingSet.input.shape[1]) * 0.6 - 0.3
# np.ones(self.trainingSet.input.shape[1])
self.activation = Activation.getActivation(activation)
self.activationPrime = Activation.getDerivative(activation)
self.activationString = activation[0].upper() + activation[1:]
self.erString = error
if error == 'absolute':
self.erf = erf.AbsoluteError()
elif error == 'different':
self.erf = erf.DifferentError()
elif error == 'mse':
self.erf = erf.MeanSquaredError()
elif error == 'sse':
self.erf = erf.SumSquaredError()
elif error == 'bce':
self.erf = erf.BinaryCrossEntropyError()
elif error == 'crossentropy':
self.erf = erf.CrossEntropyError()
else:
raise ValueError('Cannot instantiate the requested '
'error function: ' + error + 'not available')
