def __init__(self, indim, outdim, nhidden=20):
FA.__init__(self, indim, outdim)
# learning rate
self.alpha = 0.1
# number of neurons in each layer
self.indim = indim
self.nhidden = nhidden
self.outdim = outdim
# change output activation if task is classification
self.classification = False
# online training or batch, if batch, then train for that many epochs
self.online = False
self.epochs = 100
# initialize weights randomly (+1 for bias)
self.hWeights = 0.01 * np.random.random((self.nhidden, self.indim+1))
self.oWeights = 0.01 * np.random.random((self.outdim, self.nhidden+1))
# activations of neurons (sum of inputs)
self.hActivation = np.zeros((self.nhidden, 1), dtype=float)
self.oActivation = np.zeros((self.outdim, 1), dtype=float)
# outputs of neurons (after sigmoid function)
self.iOutput = np.zeros((self.indim+1, 1), dtype=float) # +1 for bias
self.hOutput = np.zeros((self.nhidden+1, 1), dtype=float) # +1 for bias
self.oOutput = np.zeros((self.outdim, 1), dtype=float)
# deltas for hidden and output layer
self.hDelta = np.zeros((self.nhidden), dtype=float)
self.oDelta = np.zeros((self.outdim), dtype=float)
def reset(self):
FA.reset(self)
# self.network = buildNetwork(self.indim, 2*(self.indim+self.outdim), self.outdim)
self.network = buildNetwork(self.indim, self.outdim, bias=True)
self.network._setParameters(random.normal(0, 0.1, self.network.params.shape))
self.pybdataset = SupervisedDataSet(self.indim, self.outdim)
def update(self, inp, tgt):
if self.online:
# train only one step online on current sample
self.predict(inp)
self.backward(tgt)
else:
# don't train, only update dataset
FA.update(self, inp, tgt)
def reset(self):
FA.reset(self)
# initialize the LWPR function
self.lwpr = LWPR(self.indim, self.outdim)
self.lwpr.init_D = 10.*np.eye(self.indim)
self.lwpr.init_alpha = 0.1*np.ones([self.indim, self.indim])
self.lwpr.meta = True
def __init__(self, indim, outdim, bayes=True, rbf=False):
""" initialize function approximator with input and output dimension. """
self.rbf = rbf
self.bayes = bayes
if self.rbf:
self.numCenters = 20
else:
self.numCenters = indim
FA.__init__(self, indim, outdim)
def reset(self):
FA.reset(self)
self.centers = [np.random.uniform(-1, 1, self.indim) for i in xrange(self.numCenters)]
self.W = np.random.random((self.numCenters, self.outdim))
# parameters for maximum map
self.alpha = 100.
self.SN = np.matrix(self.alpha*np.eye(self.numCenters))
self.mN = np.matrix(np.zeros((self.numCenters, 1), float))
def update(self, inp, tgt):
FA.update(self, inp, tgt)
inp = self._asFlatArray(inp)
tgt = self._asFlatArray(tgt)
self.pybdataset.addSample(inp, tgt)
def reset(self):
""" this initializes the function approximator to an initial state,
forgetting everything it has learned before. """
FA.reset(self)
self.matrix = np.random.uniform(-0.1, 0.1, (self.indim + 1, self.outdim))
def __init__(self, indim, outdim):
FA.__init__(self, indim, outdim)
self.filename = None
