def __init__(self, A, **kwargs):
A_hat = A.copy() + nd.eye(*A.shape)
D = nd.sum(A_hat, axis=0)
D_hat = D**-0.5
D_hat = nd.diag(D_hat)
A_hat = D_hat * A_hat * D_hat
super().__init__(A=A_hat, **kwargs)
def __init__(self,
A,
in_units,
out_units,
activation=lambda x: x,
**kwargs):
super().__init__(**kwargs)
self.in_units = in_units
self.out_units = out_units
A = A.copy() + eye(*A.shape)
with self.name_scope():
self.aggregator = MeanAggregator(A)
self.propegator = Propagator(self.in_units, self.out_units,
activation)
def __init__(self, A, in_units, out_units, activation='relu', **kwargs):
super().__init__(**kwargs)
I = nd.eye(*A.shape)
A_hat = A.copy() + I
D = nd.sum(A_hat, axis=0)
D_inv = D**-0.5
D_inv = nd.diag(D_inv)
A_hat = D_inv * A_hat * D_inv
self.in_units, self.out_units = in_units, out_units
with self.name_scope():
self.A_hat = self.params.get_constant('A_hat', A_hat)
self.W = self.params.get('W',
shape=(self.in_units, self.out_units))
if activation == 'identity':
self.activation = lambda X: X
else:
self.activation = nn.Activation(activation)
def draw(self):
'''
Draw samples from distribtions.
Returns:
`Tuple` of `mx.nd.array`s.
'''
observed_arr = nd.zeros(
(self.__batch_size, 1, self.__cluster_n, self.__cluster_n),
ctx=self.__ctx)
observed_arr[:, 0] = nd.eye(self.__cluster_n, ctx=self.__ctx)
observed_arr = observed_arr + nd.random.uniform(
low=self.__low,
high=self.__high,
shape=(self.__batch_size, 1, self.__cluster_n, self.__cluster_n),
ctx=self.__ctx)
if self.noise_sampler is not None:
observed_arr = observed_arr + self.noise_sampler.draw()
return observed_arr
model.add(features)
logger.info("GCN Summary: \n{}".format(model))
classifier = LogisticRegressor(out_units)
model.add(classifier)
logger.info("GCN + LR Summary: \n{}".format(model))
model.hybridize()
model.initialize(Uniform(1))
return model, features
# ## Model 1: Identity Matrix as Features
X_1 = I = nd.eye(*A.shape)
model_1, features_1 = build_model(A, X_1)
model_1(X_1)
# ## Model 2: Distance to Administrator and Instructor as Additional Features
X_2 = nd.zeros((A.shape[0], 2))
node_distance_instructor = shortest_path_length(zkc.network, target=33)
node_distance_administrator = shortest_path_length(zkc.network, target=0)
for node in zkc.network.nodes():
X_2[node][0] = node_distance_administrator[node]
X_2[node][1] = node_distance_instructor[node]
X_2 = nd.concat(X_1, X_2)
model_2, features_2 = build_model(A, X_2)
