torch_geometric_signed_directed.nn.directed.DIMPA
=================================================

.. py:module:: torch_geometric_signed_directed.nn.directed.DIMPA


Classes
-------

.. autoapisummary::

   torch_geometric_signed_directed.nn.directed.DIMPA.DIMPA


Module Contents
---------------

.. py:class:: DIMPA(hop: int, fill_value: float = 0.5)

   Bases: :py:obj:`torch.nn.Module`


   The directed mixed-path aggregation model from the
   `DIGRAC: Digraph Clustering Based on Flow Imbalance <https://proceedings.mlr.press/v198/he22b.html>`_ paper.

   :Parameters: * **hop** (*int*) -- Number of hops to consider.
                * **fill_value** (*float, optional*) -- The layer computes
                  :math:`\mathbf{\hat{A}}` as :math:`\mathbf{A} + fill_value*\mathbf{I}`.
                  (default: :obj:`0.5`)


   .. py:attribute:: conv_layer


   .. py:method:: forward(x_s: torch.FloatTensor, x_t: torch.FloatTensor, edge_index: torch.FloatTensor, edge_weight: torch.FloatTensor) -> torch.FloatTensor

      Making a forward pass of DIMPA.

      Arg types:
          * **x_s** (PyTorch FloatTensor) - Souce hidden representations.
          * **x_t** (PyTorch FloatTensor) - Target hidden representations.
          * **edge_index** (PyTorch FloatTensor) - Edge indices.
          * **edge_weight** (PyTorch FloatTensor) - Edge weights.
      Return types:
          * **feat** (PyTorch FloatTensor) - Embedding matrix, with shape (num_nodes, 2*input_dim).