add unisage

test/nn/hypergraph/test_unisage.py

@@ -0,0 +1,25 @@
+"""Test the UniGIN class."""
+
+import numpy as np
+import torch
+
+from topomodelx.nn.hypergraph.unisage import UniSAGE
+
+
+class TestUniGIN:
+    """Test the UniGIN."""
+
+    def test_fowared(self):
+        """Test forward method."""
+        device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+
+        incidence = torch.from_numpy(np.random.rand(2, 2)).to_sparse_csr()
+        incidence = incidence.float().to(device)
+        model = UniSAGE(channels_edge=2, channels_node=2, n_layers=2).to(device)
+        x_0 = torch.rand(2, 2)
+
+        x_0 = torch.tensor(x_0).float().to(device)
+
+        y1 = model(x_0, incidence)
+
+        assert y1.shape == torch.Size([2])

topomodelx/nn/hypergraph/unisage.py

@@ -0,0 +1,57 @@
+"""UniSAGE class."""
+
+import torch
+
+from topomodelx.nn.hypergraph.unisage_layer import UniSAGELayer
+
+
+class UniSAGE(torch.nn.Module):
+    """Neural network implementation of UniSAGE for hypergraph classification.
+
+    Parameters
+    ----------
+    channels_edge : int
+        Dimension of edge features
+    channels_node : int
+        Dimension of node features
+    n_layer : 2
+        Amount of message passing layers.
+
+    """
+
+    def __init__(self, channels_edge, channels_node, n_layers=2):
+        super().__init__()
+        layers = []
+        for _ in range(n_layers):
+            layers.append(
+                UniSAGELayer(
+                    in_channels=channels_edge,
+                    out_channels=channels_edge,
+                )
+            )
+        self.layers = torch.nn.ModuleList(layers)
+        self.linear = torch.nn.Linear(channels_edge, 1)
+
+    def forward(self, x_1, incidence_1):
+        """Forward computation through layers, then linear layer, then global max pooling.
+
+        Parameters
+        ----------
+        x_1 : tensor
+            shape = [n_edges, channels_edge]
+            Edge features.
+
+        incidence_1 : tensor
+            shape = [n_nodes, n_edges]
+            Boundary matrix of rank 1.
+
+        Returns
+        -------
+        _ : tensor
+            shape = [1]
+            Label assigned to whole complex.
+        """
+        for layer in self.layers:
+            x_1 = layer(x_1, incidence_1)
+        pooled_x = torch.max(x_1, dim=0)[0]
+        return torch.sigmoid(self.linear(pooled_x))