fix tensor indices bug

predictor/model/models.py

 
 class Connector(nn.Module):
     def __init__(self, gpu_id=None):
+        self.gpu_id = gpu_id
         super(Connector, self).__init__()
 
         self.ddiDataset = DDInteractionDataset()
         x = self.gcn(x, edge_index)
         drug1_idx = torch.flatten(drug1_idx)
         drug2_idx = torch.flatten(drug2_idx)
-        drug1_feat = x[drug1_idx]
-        drug2_feat = x[drug2_idx]
+        #drug1_feat = x[drug1_idx]
+        #drug2_feat = x[drug2_idx]
+        drug1_feat = torch.empty((len(drug1_idx), len(x[0])))
+        drug2_feat = torch.empty((len(drug2_idx), len(x[0])))
+        for index, element in enumerate(drug1_idx):
+            drug1_feat[index] = (x[element])
+        for index, element in enumerate(drug2_idx):
+            drug2_feat[index] = (x[element])
+        if self.gpu_id is not None:
+            drug1_feat = drug1_feat.cuda(self.gpu_id)
+            drug2_feat = drug2_feat.cuda(self.gpu_id)
         for i, x in enumerate(drug1_idx):
             if x < 0:
                 drug1_feat[i] = get_FP_by_negative_index(x)