def __init__(self,
inputs,
outputs,
numHiddenNeurons,
activationFunction,
LN=True,
AE=True,
ORTH=True,
inputWeightForgettingFactor=0.999,
outputWeightForgettingFactor=0.999,
hiddenWeightForgettingFactor=0.999):
self.activationFunction = activationFunction
self.inputs = inputs
self.outputs = outputs
self.numHiddenNeurons = numHiddenNeurons
# input to hidden weights
self.inputWeights = np.random.random(
(self.numHiddenNeurons, self.inputs))
# hidden layer to hidden layer wieghts
self.hiddenWeights = np.random.random(
(self.numHiddenNeurons, self.numHiddenNeurons))
# initial hidden layer activation
self.initial_H = np.random.random((1, self.numHiddenNeurons)) * 2 - 1
self.H = self.initial_H
self.LN = LN
self.AE = AE
self.ORTH = ORTH
# bias of hidden units
self.bias = np.random.random((1, self.numHiddenNeurons)) * 2 - 1
# hidden to output layer connection
self.beta = np.random.random((self.numHiddenNeurons, self.outputs))
# auxiliary matrix used for sequential learning
self.M = inv(0.00001 * np.eye(self.numHiddenNeurons))
self.forgettingFactor = outputWeightForgettingFactor
self.trace = 0
self.thresReset = 0.001
if self.AE:
self.inputAE = FOSELM(inputs=inputs,
outputs=inputs,
numHiddenNeurons=numHiddenNeurons,
activationFunction=activationFunction,
LN=LN,
forgettingFactor=inputWeightForgettingFactor,
ORTH=ORTH)
self.hiddenAE = FOSELM(inputs=numHiddenNeurons,
outputs=numHiddenNeurons,
numHiddenNeurons=numHiddenNeurons,
activationFunction=activationFunction,
LN=LN,
ORTH=ORTH)
def __init__(self, inputs, outputs, numHiddenNeurons, activationFunction, min_t, max_t, LN=True, AE=True, ORTH=True,
inputWeightForgettingFactor=0.999,
outputWeightForgettingFactor=0.999,
hiddenWeightForgettingFactor=0.999):
self.min_t = min_t
self.max_t = max_t
self.activationFunction = activationFunction
self.inputs = inputs
self.outputs = outputs
self.numHiddenNeurons = numHiddenNeurons
# input to hidden weights
self.inputWeights = np.random.random((self.numHiddenNeurons, self.inputs))
# hidden layer to hidden layer weights
self.hiddenWeights = np.random.random((self.numHiddenNeurons, self.numHiddenNeurons))
# initial hidden layer activation
self.initial_H = np.random.random((1, self.numHiddenNeurons)) * 2 - 1
self.H = self.initial_H
self.LN = LN
self.AE = AE
self.ORTH = ORTH
# bias of hidden units
self.bias = np.random.random((1, self.numHiddenNeurons)) * 2 - 1
self.forgettingFactor = outputWeightForgettingFactor
self.FFmin = 0.9
self.FFmax = 0.999
self.trace = 0
self.thresReset = 0.001
if self.AE:
self.inputAE = FOSELM(inputs=inputs,
outputs=inputs,
numHiddenNeurons=numHiddenNeurons,
activationFunction=activationFunction,
LN=LN,
forgettingFactor=inputWeightForgettingFactor,
ORTH=ORTH
)
self.hiddenAE = FOSELM(inputs=numHiddenNeurons,
outputs=numHiddenNeurons,
numHiddenNeurons=numHiddenNeurons,
activationFunction=activationFunction,
LN=LN,
ORTH=ORTH
)
def __init__(self, inputs, outputs, numHiddenNeurons, activationFunction, LN=True,
outputWeightForgettingFactor=0.999,
inputWeightForgettingFactor=0.999, AE=True, ORTH=False):
self.activationFunction = activationFunction
self.inputs = inputs
self.outputs = outputs
self.numHiddenNeurons = numHiddenNeurons
# input to hidden weights
self.inputWeights = np.random.random((self.numHiddenNeurons, self.inputs))
# bias of hidden units
#self.bias = np.random.random((1, self.numHiddenNeurons)) * 2 - 1
self.bias = np.zeros([1,self.numHiddenNeurons])
# hidden to output layer connection
self.beta = np.random.random((self.numHiddenNeurons, self.outputs))
# auxiliary matrix used for sequential learning
self.M = None
self.LN = LN
self.AE = AE
self.ORTH = ORTH
self.forgettingFactor =outputWeightForgettingFactor
self.FOSELM_AE = FOSELM(inputs= inputs,
outputs = inputs,
numHiddenNeurons= numHiddenNeurons,
activationFunction= activationFunction,
LN=True,
forgettingFactor=inputWeightForgettingFactor,
ORTH=self.ORTH)