How to pass edge_index to LinkNeighborLoader?

[songsong0425]

Hi, I have a question about the LinkNeighborLoader.
In the previous question, I solved my problem by passing edge_label_index to LinkNeighborLoader synchronizing to the original dataset.
But I found that the loader initializes the edge_index based on the edge_label_index.

train_data
# Data(edge_index=[2, 2159112], pos_edge_label=[3955], pos_edge_label_index=[2, 3955], edge_label=[3955, 5], edge_label_index=[2, 3955], x=[47822, 512])

train_loader = LinkNeighborLoader(train_data, edge_index = train_data.edge_index, edge_label = train_data.edge_label, edge_label_index=train_data.edge_label_index, batch_size=32, shuffle=True, neg_sampling_ratio=None, num_neighbors=[3,3])
for batch in val_loader:
    print(batch)
# Data(edge_index=[2, 513], pos_edge_label=[564], pos_edge_label_index=[2, 564], edge_label=[32, 5], edge_label_index=[2, 32], x=[366, 512], n_id=[366], e_id=[513], num_sampled_nodes=[3], num_sampled_edges=[2], input_id=[32])
# Data(edge_index=[2, 492], pos_edge_label=[564], pos_edge_label_index=[2, 564], edge_label=[32, 5], edge_label_index=[2, 32], x=[358, 512], n_id=[358], e_id=[492], num_sampled_nodes=[3], num_sampled_edges=[2], input_id=[32])
# ...

But I need the original edge_index in train_data since I want to predict the link in the entire knowledge graph.
So I tried passing the edge_inex parameter into the loader and it returned the error below:

train_loader = LinkNeighborLoader(train_data, edge_index = train_data.edge_index, edge_label = train_data.edge_label, edge_label_index=train_data.edge_label_index, batch_size=32, shuffle=True, neg_sampling_ratio=None, num_neighbors=[3,3])

TypeError: init() got an unexpected keyword argument 'edge_index'

Actually, I need to perform the message passing in the entire knowledge graph (i.e., edge_index=[2, 2159112]) and predict the link between nodes in the training/validation/test dataset. Isn't edge_index related to this task?

[songsong0425]

Now I understand that edge_index for the message passing is used in the deep learning model like model(data.x, data.edge_index) and edge_index in Loader is automatically allocated by sampled nodes converting their own ids. Although I close this comment by solving the question, please comment to me if there is any wrong information in my opinion. Thanks!