CogDL v0.2.0

A new major release!! It includes easy-to-use experiment and pipeline APIs for all experiments and applications. It also provides oagbert API. Thanks to all the contributors 🎉

New Features

• #142 Add experiment API (see examples/quick_start.py for reference)
• #151 Enable automl feature in experiment API, the usage is in README
• #157 Add pipeline API (see examples/pipeline.py for reference)
• #153 Add oagbert API (see examples/oagbert.py for reference)
• #59 Add similarity search task
• #78 Add neighborhood sampler for large-scale training
• #87 Apply GNN for link prediction task

New Models

• #67 Add SGC model (thanks to @khtee)
• #60 Add SGC-PN model (thanks to @feng-y16)
• #63 Add PPNP model (thanks to @TiagoMAntunes)
• #68 Add SAGPool model (thanks to @frouioui)
• #69 Add GDC_GCN model (thanks to @kwyoke)
• #74 Add JKNet (thanks to @WXR1998)
• #76 Add SIGN model (thanks to @hmartelb)
• #80 Add HGP-SL model (thanks to @Sahandfer)
• #88 Add DropEdge model (thanks to @JiaYiLiJayee)
• #96 Add Graph U-Net model
• #102 Add PPRGo model

New Datasets

• #158 Add Yelp/Amazon datasets in this paper.

Bug Fixes

• #141 Fix bugs when using CPU

Requirement Update

• CogDL now requires numba
• CogDL now requires transformers

Document Update

• #140 Update the structure of the document
• #143~#147 Fix readthedocs build

Miscellaneous

• #61 Introduce Code style (thanks to @MaLiN2223)
• #66 Create dockerfile for CogDL (thanks to @TiagoMAntunes)
• #86 Add a script for contributing a new model (thanks to @Sahandfer)
• #133 Add templates for github issues and pull requests
• #135 Integrate the training and evaluation of self-supervised models with a trainer

CogDL v0.1.2

New Features

• #48 Support the pre-training task on molecule graphs
• #38 Add Trainer API for flexible training
• #38 Add Sampler API for training large-scale datasets and now supports GraphSAINT sampler.

New Models

• #48 STP-GNN for pre-training
• #38 GPT-GNN for node classification
• #39 Triple based knowledge embedding methods (complex, distmult, rotate, transe)
• #48 DeeperGCN for node classification
• #48 GCNII for node classification

New Results

• #51 Update the leaderboard of the unsupervised node classification task
• #48 Update the leaderboard of the semi-supervised node classification task
• #48 Update the leaderboard of the graph classification task

New Datasets

• #50 Add some molecule datasets
  • "bio" and "chem" in Jure's paper.
  • BBBP and BACE
• #38 Add OGB datasets

New Examples

• #51 Add many examples of embedding methods
• #48 Add many examples of graph neural networks

Requirement Update

• #38 CogDL now requires ogb

Miscellaneous

• #50 #54 Remove saved/ folder and support downloading pre-trained GCC model
• #52 Improve the coverage to 80%

CogDL v0.1.1

New Features

• Support link prediction task on knowledge graphs
• Support hyper-parameter search using optuna

New Models

• GCC for graph classification: GCC is a contrastive learning framework that implements unsupervised structural graph representation pre-training.
• GRAND for node classification (thanks to @wzfhaha): GRAND randomly drops node features in training process to implement data augmentatoin and achieves sota in benchmarks.
• DGI for unsupervised node classification: DGI applies local-global contrastive learning methods to train GNN and first achieves results comparable to semi-supervised methods in benchmarks.
• MVGRL for unsupervised node classification: MVGRL is a self-supervised approach based on contrastive multi-view learning to learn representations.
• ProNE++ for unsupervised node classification: ProNE++ employs graph filter and AutoML to help enhance node embeddings.
• GraphSAGE for unsupervised node classification: unsupervised version of GraphSAGE.
• DisenGCN for node classification: DisenGCN disentangles node representations by separating different factors.
• CompGCN/RGCN for KG link prediction: RGCN and CompGCN are GNNs for knowledge graph embedding considering the type of edges.

New Results

• GCC results for heterogeneous node classification task

New Datasets

• Add some knowledge graph datasets (FB/WN datasets)

New Examples

• Add an example using hyper-parameter search
• Add an example using a custom dataset/model

Bug Fixes

• Fixed "division by zero" bug in Sparse GAT model

Requirement Update

• CogDL now requires optuna
• CogDL does not require dgl.model_zoo anymore.

Miscellaneous

• Add a check whether tuples of (task, model, dataset) are matching in the training script
• Add a GCC pre-trained model in saved/

CogDL v0.1.0

The first open release includes basically everything in the repository.

• Basic CogDL APIs and systems
• Use PyTorch backend
• Design several important graph tasks
• Implement lots of models based on PyTorch and PyTorch Geometric
• Support running by the command line interface
• Provide leaderboards for tasks
• Provide basic tutorials and documents