Need Some Help in Message_Passing Class

[utkarsh0902311047]

In the message passing class, there is a propagate function. I want to collect all the results of "_collect," especially pos[i] , pos[j], edge_index in a file every time it goes through it doesn't matter how many epochs or how many times it goes through _collect in an epoch. I want to collect all these values in a file in readable format.
How can I do it?

[rusty1s]

You can leverage hooks for this, see https://pytorch-geometric.readthedocs.io/en/latest/generated/torch_geometric.nn.conv.MessagePassing.html#torch_geometric.nn.conv.MessagePassing.register_propagate_forward_pre_hook.

[rusty1s]

You need to register on the instantiated object, not the class. So it would be something like self.conv.register_propagate_forward_pre_hook(hook).

[utkarsh0902311047]

NameError: name 'self' is not defined

I am getting this error

[rusty1s]

Depends on where your layers are defined. How does your model look like?

[utkarsh0902311047]

The model is pointnet2_classification.py

[rusty1s]

Then this would be

model.sa1_module.conv.register_propagate_forward_pre_hook(...)