Model: HERO ADDED

openhgnn/config.ini

@@ -1559,3 +1559,24 @@ mini_batch_flag = True
 emb_dim=20
 hid_dim=64
 batch_size=128
+
+[HERO]
+lr = 0.0025
+patience = 60
+nb_epochs = 500
+hid_units = 256
+hid_units2 = 128
+out_ft = 256
+g_dim = 64
+g_equidim = 256
+p_equidim = 256
+edge_rate = 0.9
+alpha = 100
+beta = 500
+gamma = 10
+eta = 1
+lambbda = 1
+dataset = 'acm4GTN'
+gpu = 1
+mode = 0
+save_dir = './saved_models'

openhgnn/layers/Fully_connect.py

@@ -0,0 +1,55 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+import numpy as np
+
+class FullyConnect(nn.Module):
+    def __init__(self, in_ft, out_ft, act=nn.Identity(), drop_prob=0.0, isBias=False):
+        super().__init__()
+        self.fc = nn.Linear(in_ft, out_ft, bias=False)
+        if self.fc.bias is not None:
+            self.fc.bias.data.fill_(0.0)
+
+        if isBias:
+            self.bias = nn.Parameter(torch.empty(out_ft))
+            self.bias.data.fill_(0.0)
+        else:
+            self.register_parameter('bias', None)
+
+        self.act = act
+        self.drop_prob = drop_prob
+        self.isBias = isBias
+
+        for m in self.modules():
+            self.weights_init(m)
+
+    def weights_init(self, m):
+        if isinstance(m, nn.Linear):
+            torch.nn.init.xavier_uniform_(m.weight.data)
+            if m.bias is not None:
+                m.bias.data.fill_(0.0)
+
+
+    def forward(self, emb):
+        # emb (batch_size, ft)
+        emb = F.dropout(emb, self.drop_prob, training=self.training)
+        e = self.fc(emb)  #  (batch_size, d)
+        if self.isBias:
+            e += self.bias
+        return self.act(e)
+
+
+class FullyConnect2(nn.Module):
+    def __init__(self, in_ft, hid_unit, out_ft, drop_prob=0.0, isBias=False):
+        super().__init__()
+        self.fc = FullyConnect(in_ft, hid_unit, act=nn.PReLU(), drop_prob=drop_prob,isBias=isBias)
+        self.fc2 = FullyConnect(hid_unit, out_ft, act=nn.PReLU(), drop_prob=drop_prob,isBias=isBias)
+        self.dense = FullyConnect(out_ft, 1, act=nn.Identity(), drop_prob=drop_prob,isBias=isBias)
+
+
+    def forward(self, emb):
+        # emb (batch_size, ft)
+        e = self.fc(emb)  #  (batch_size, d)
+        e2 = self.fc2(e)
+        out = self.dense(e2)
+        return out

openhgnn/layers/Linear_layer.py

@@ -0,0 +1,52 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+import numpy as np
+
+class Linear_layer(nn.Module):
+    def __init__(self, in_ft, out_ft, act=nn.PReLU(), drop_prob=0.0, isBias=False):
+        super().__init__()
+        self.linear = nn.Linear(in_ft, out_ft, bias=False)
+
+#         if isBias:
+#             self.bias = nn.Parameter(torch.empty(out_ft))
+#             self.bias.data.fill_(0.0)
+#         else:
+#             self.register_parameter('bias', None)
+
+        self.act = act
+        self.isBias = isBias
+        self.drop_prob = drop_prob
+
+        for m in self.modules():
+            self.weights_init(m)
+
+    def weights_init(self, m):
+        if isinstance(m, nn.Linear):
+            torch.nn.init.xavier_uniform_(m.weight.data)
+            if m.bias is not None:
+                m.bias.data.fill_(0.0)
+
+    def forward(self, emb):
+        # emb (batch_size, ft)
+#         emb = F.dropout(emb, self.drop_prob, training=self.training)
+        e = self.linear(emb) #  (batch_size, d)
+#         if self.isBias:
+#             e += self.bias
+        e_out = self.act(e)
+        return e_out
+
+
+class GNN(nn.Module):
+    def __init__(self, nb_rel, in_ft, out_ft, act=nn.PReLU(), drop_prob=0.5, isBias=False):
+        super().__init__()
+        self.encoder = nn.ModuleList()
+        for i in range(nb_rel):
+            self.encoder.append(GCN(in_ft, out_ft, act=act, isBias=isBias))
+
+    def forward(self, embs):
+        outs = []
+        for emb in embs:  # emb (batch_size, ft)
+            outs.append(self.encoder(emb))
+        outs = torch.stack(outs, 0)  # outs (nb_rel, batch_size, ft)
+        return outs

openhgnn/layers/__init__.py

@@ -8,7 +8,8 @@
 from .AdapropT import *
 from .AdapropI import *
 from .rgcn_layer import *
-
+from .Fully_connect import *
+from .Linear_layer import *
 __all__ = [
     'HeteroEmbedLayer',
     'GeneralLinear',
@@ -24,7 +25,9 @@
     'AttConv',
     'LSTMConv',
     'AdapropT',
-    'AdapropI'
+    'AdapropI',
+    'Fully_connect',
+    'Linear_layer',
 ]
 
 classes = __all__