Version Update for TGB

Major Update: Please update your package to 2.0.0 (June 11th, 2024).
  • added four temporal knowledge graph datasets and four temporal heterogeneous graph datasets.
  • provided leaderboard and examples for these datasets
  • the preprint for TGB 2.0 is available.
package version 0.9.2** (December 5th, 2023).
  • fix the time conversion error in tgbl-flight, now the edgelist directly contain the converted unix timestamps.
  • if you have older download of this datasets, please remove TGB/tgb/datasets/tgbl_flight folder and do a clean redownload.
  • no negative samples or dataset edges are changed, only now that the provided edges now have unix timestamps instead of date.
package version 0.9.0 (September 27th, 2023).
  • add a novel large scale dataset tgbn-token with 72 million edges.
  • leaderboard for tgbn-token is added to tgbn leaderboards.
  • fixed an error where a small percentage of edges in tgbl-flight and tgbl-coin is not sorted chronologically. Updated tgbl-flight and tgbl-coin to version 2.
  • tgbl-flight and tgbl-coin version 2 leaderboards are updated accordingly.
package version 0.8.0 (August 6th, 2023).
  • addressed issues pointed out by the authors in a recent paper, thanks for the suggestions! Happy to incorporate more community feedbacks in the future.
  • addressed metric computation error in the nodeproppred task, specially computing the average over the number of label batches now, please refer to the examples/nodeproppred/ folder to see corrected metric computations.
  • no longer applys log softmax at the NodePredictor, no significant numerical difference between applying or not
  • update to version 2 of tgbn leaderboards with the updated metric computation.
package version 0.7.5 (July 27th, 2023).
  • updated tgbl-wiki and tgbl-review to version 2, the number of negative samples have been increased for both datasets
  • tgbl-wiki and tgbl-review leaderboards have been updated accordingly
  • added leaderboard submission form
package version 0.7.0 (July 21th, 2023).
  • fixed inconsistency between dataset.edge_feat and dataset.full_data['edge_feat']
  • fixed off by one batch error in tgb.nodeproppred.dataset.py
  • thanks to the feedbacks from the community
package version 0.6.0 (June 30th, 2023).
  • initial release of TGB
package version 0.1.2 (June 12th, 2023).