def pretrain(seed=None, data=None, topology=None, nepoch=50, weights=None):
if weights is None:
weights = [] # initialize weights array
if len(topology) < 2: # we're done
return weights
print("\n[Train] Layer {}/{}, Nr of Epoch: {}".format(len(weights)+1, len(weights)+len(topology)-1, nepoch))
# create diabolo network (autoencoder)
dn = DN([topology[0], topology[1], topology[0]], seed)
dn.init()
# create new dataset
dataset = SimpleContainer(data, seed)
dataset.init()
# train DN on new dataset using backpropagation
run_program(dataset, dn, bp, nepoch)
# for each instance, calculate ilp map of middle layer
xlayer = np.array([dn.compute_ilp_map([instance])[1] for instance in dataset.data['x']], dtype=np.float16)
# concat newly trained weight matrix of input to hidden layer to stored weights
weights.append(dn.weights[0])
# next layer
pretrain(seed, {'x': xlayer, 'y': xlayer}, topology[1:], nepoch, weights)
return weights
def load_dataset(self, seed=None):
rdfGraph = rdf.KnowledgeGraph()
rdfGraph.read("http://www.w3.org/People/Berners-Lee/card")
dataset = SimpleContainer(rdf.propositionalize(rdfGraph, twoHot), seed)
dataset.init()
return dataset
def load_dataset(self, seed=None):
rdfGraph = rdf.KnowledgeGraph()
rdfGraph.read("./examples/das.ttl")
dataset = SimpleContainer(rdf.propositionalize(rdfGraph, twoHot), seed)
dataset.init()
return dataset
def load_dataset(self, seed=None):
arffFile = arff.ARFF('./examples/iris.arff')
arff.toNumerical(arffFile)
arffFile.data = aux.normalize(arffFile.data)
dataset = SimpleContainer(arff.toVector(arffFile.data, range(0, 4), range(4, 5)), seed)
dataset.init()
return dataset
