def fit(self, XTrain=None, yTrain=None, KTrain=None, delta=None, folds=0):
self.nTrain = yTrain.shape[0]
self.yTrain = yTrain
if KTrain is not None:
self.kernel = KTrain
self.S, self.U = LA.eigh(self.kernel + SP.eye(self.nTrain) * 1e-8)
self.Uy = SP.dot(self.U.T, yTrain)
self.Uone = SP.dot(self.U.T, SP.ones_like(yTrain))
if delta is None:
# Initialize cross-learning of delta
if folds > 0:
self.ldelta = self.crossvalidate_delta(folds)
else:
self.ldelta = dlimix.optdelta(self.Uy[:, 0], self.Uone, self.S,
100000, -5.0, 5.0)
self.delta = SP.exp(self.ldelta)
else:
self.delta = delta
self.ldelta = SP.log(delta)
self.cov = LA.inv((self.kernel + SP.eye(self.nTrain) * self.delta))
def fit(self, XTrain=None, yTrain=None, KTrain=None, delta=None, folds=0):
self.nTrain = yTrain.shape[0]
self.yTrain = yTrain
if KTrain is not None:
self.kernel = KTrain
self.S, self.U = LA.eigh(self.kernel+SP.eye(self.nTrain)*1e-8)
self.Uy = SP.dot(self.U.T, yTrain)
self.Uone = SP.dot(self.U.T, SP.ones_like(yTrain))
if delta is None:
# Initialize cross-learning of delta
if folds > 0:
self.ldelta = self.crossvalidate_delta(folds)
else:
self.ldelta = dlimix.optdelta(self.Uy[:, 0], self.Uone, self.S,
100000, -5.0, 5.0)
self.delta = SP.exp(self.ldelta)
else:
self.delta = delta
self.ldelta = SP.log(delta)
self.cov = LA.inv((self.kernel + SP.eye(self.nTrain)*self.delta))
def update_delta(self, mean):
Umean = SP.dot(self.U.T, mean)
self.ldelta = dlimix.optdelta(self.Uy[:, 0], Umean, self.S, 100,
self.ldelta - 1., self.delta + 1.)
self.delta = SP.exp(self.ldelta)
return self.delta
def update_delta(self, mean):
Umean = SP.dot(self.U.T, mean)
self.ldelta = dlimix.optdelta(self.Uy[:, 0], Umean, self.S, 100,
self.ldelta-1., self.delta+1.)
self.delta = SP.exp(self.ldelta)
return self.delta