def set_data(self, data, ignore_feat=True):
"""
:param data: a TxKxV tensor of event-feature map where T- time; K- process id; V- corresponding feature vector
:input feature: V+1 features, [V features, 1 bias], bias- denotes recency
"""
assert isinstance(
data, np.ndarray
) and data.ndim == 3, "data needs to be a TxKxV tensor of event features"
self.update_K(data.shape[1])
self.F = data.shape[2] - 1
# calculate linear mark function: F expected to be KxKxTxF tensor
if self.F < 1 or ignore_feat:
f_mat = None
self.F = 0
else:
f_mat = data[:, :, :-1]
if self.F > 0:
self.v = 1e-3 * np.ones((self.K, self.F))
self.win = np.array([
self.basis(i / float(self.dt), **self.basis_args)
for i in xrange(data.shape[0])
])
self.data = data # orignal event features, shape: TxKxF
# convolve data with windowed signal
self.conv_data = np.concatenate([
convolve(data[:, k, -1], self.win)[:, None]
for k in xrange(data.shape[1])
],
axis=1)
self.conv_data_fea = f_mat # convolved to feature matrix, shape: TxKxF
def set_data(self, data, dyadic_data, ignore_feat=True):
"""
:param dyadic_data: expect dyadic interaction data between processes, shape: KxKxT
"""
assert isinstance(dyadic_data,
np.ndarray) and data.ndim == 3, "dyadic_data needs be a KxKxT tensor of interaction events"
super(LinearDiscreteHawkes, self).set_data(data, ignore_feat)
self.conv_dyad_data = dyadic_data
# convolution step
for k1 in xrange(self.conv_dyad_data.shape[0]):
for k2 in xrange(self.conv_dyad_data.shape[1]):
self.conv_dyad_data[k1, k2, :] = self.conv_dyad_data[k1, k2, :] + convolve(self.conv_dyad_data[k1, k2, :], self.win)
def set_data(self, data, dyadic_data, ignore_feat=True):
"""
:param dyadic_data: expect dyadic interaction data between processes, shape: KxKxT
"""
assert isinstance(
dyadic_data, np.ndarray
) and data.ndim == 3, "dyadic_data needs be a KxKxT tensor of interaction events"
super(LinearDiscreteHawkes, self).set_data(data, ignore_feat)
self.conv_dyad_data = dyadic_data
# convolution step
for k1 in xrange(self.conv_dyad_data.shape[0]):
for k2 in xrange(self.conv_dyad_data.shape[1]):
self.conv_dyad_data[
k1, k2, :] = self.conv_dyad_data[k1, k2, :] + convolve(
self.conv_dyad_data[k1, k2, :], self.win)
def set_data(self, data, ignore_feat=True):
"""
:param data: a TxKxV tensor of event-feature map where T- time; K- process id; V- corresponding feature vector
:input feature: V+1 features, [V features, 1 bias], bias- denotes recency
"""
assert isinstance(data, np.ndarray) and data.ndim == 3, "data needs to be a TxKxV tensor of event features"
self.update_K(data.shape[1])
self.F = data.shape[2]-1
# calculate linear mark function: F expected to be KxKxTxF tensor
if self.F < 1 or ignore_feat:
f_mat = None
self.F = 0
else:
f_mat = data[:, :, :-1]
if self.F > 0:
self.v = 1e-3 * np.ones((self.K, self.F))
self.win = np.array([self.basis(i/float(self.dt), **self.basis_args) for i in xrange(data.shape[0])])
self.data = data # orignal event features, shape: TxKxF
# convolve data with windowed signal
self.conv_data = np.concatenate([convolve(data[:, k, -1], self.win)[:, None] for k in xrange(data.shape[1])], axis=1)
self.conv_data_fea = f_mat # convolved to feature matrix, shape: TxKxF
