In current implementation for nodeproppred task
- last batch of edges from a day is split based on which edges are in previous day and which edges are in new day
- iterate through edges in the previous day
- predict for the batch of node labels in the next day
- backprop for the above batch of node labels
- iterate through the edges in new day
issue: the above setup doesn't support the case where there are edges from multiple days in a single batch
currently, the batch size is small thus each batch contains edges from 1 day mostly.
In current implementation for
nodeproppredtaskissue: the above setup doesn't support the case where there are edges from multiple days in a single batch
currently, the batch size is small thus each batch contains edges from 1 day mostly.