torch.nn.modules.module.ModuleAttributeError: 'GATConvNoSelfLoops' object has no attribute 'weight'

[cravisjan97]

I am running training_job_example_aliens_superpix.py and I am getting the above attribute error during forward() in model file

[lidan233]

I meet the same situation. And I find your code in forward method is not same as GATConv's forward method, which is a contradiction for your statements.
# The part before the feature propagation is exactly the same as in the # parent class GATConv, since the primal features get multiplied by the # weight matrix. However, self-loops are not added, as the would not # have a corresponding node in the dual graph.