def __init__(self, adj, input_dim: int, output_dim: int, **kwargs):
super(GCN, self).__init__()
self.register_buffer(
'laplacian',
calculate_laplacian_with_self_loop(torch.FloatTensor(adj)))
self._num_nodes = adj.shape[0]
self._input_dim = input_dim # seq_len for prediction
self._output_dim = output_dim # hidden_dim for prediction
self.weights = nn.Parameter(
torch.FloatTensor(self._input_dim, self._output_dim))
self.reset_parameters()
def __init__(self,
adj,
num_gru_units: int,
output_dim: int,
bias: float = 0.0):
super(TGCNGraphConvolution, self).__init__()
self._num_gru_units = num_gru_units
self._output_dim = output_dim
self._bias_init_value = bias
self.register_buffer(
'laplacian',
calculate_laplacian_with_self_loop(torch.FloatTensor(adj)))
self.weights = nn.Parameter(
torch.FloatTensor(self._num_gru_units + 1, self._output_dim))
self.biases = nn.Parameter(torch.FloatTensor(self._output_dim))
self.reset_parameters()