MultiGPrompt

We provide the code (in pytorch) and datasets for our paper "MultiGPrompt for Multi-Task Pre-Training and Prompting on Graphs", which is accepted by WWW2024.

Description

The repository is organised as follows:

• data/: contains data we use.
• modelset/: contains pre-trained model we use.
• MutilGPrompt_CoraCiteseer_node/: implements pre-training and node level downstream tasks for Cora and Citeseer.
• MutilGPrompt_TU_node/: implements pre-training and node level downstream task for ENZYMES and PROTEINS.
• MutilGPrompt_TU_graph/: implements pre-training for BZR and COX2, graph level downstream task for BZR,COX2,ENZYMES,PROTEINS.

Package Dependencies

• python 3.8.16
• pytorch 1.10.1
• cuda 11.3
• pyG 2.0.0

Running experiments

Due to the limitation of file size, we upload all datasets except for PROTEINS. But the pre-trained model for each datasets is uploaded.

Node Classification for Cora and Citeseer

Default dataset is Cora. You need to change the corresponding parameters in preprompt.py, downprompt.py.py and execute.py to train and evaluate on other datasets.

Pretrain and Prompt tune: python execute.py

Node Classification for ENZYMES and PROTEINS

Default dataset is ENZYMES. You need to change the corresponding parameters in preprompt.py, downprompt.py.py and execute.py to train and evaluate on other datasets.

Pretrain and Prompt tune: python execute.py

Graph Classification for BZR,COX2,ENZYMES,PROTEINS.

Default dataset is ENZYMES. You need to change the corresponding parameters in preprompt.py, downprompt.py.py and execute.py to train and evaluate on other datasets.

Pretrain and Prompt tune: python execute.py