Requirements

torch==2.4.1  
dgl-cu113==0.9.0  
diffusers==0.35.1
huggingface-hub==0.35.0

Environment

• Python: Python 3.8
• CUDA: CUDA 11.8
• GPU: A800-80GB 80GB Memory

Datasets and Best_model

Datasets and best_model files are at the releases of this repository.

Training

python main.py --dataset ICEWS14-few --data_path ./ICEWS14-few --few 5 --data_form Pre-Train --prefix icews14_frecd_iter100_lr1e-3_margin1 --device 0 --batch_size 256 --g_batch 1024 -epo 10000 --eval_epoch 100 --checkpoint_epoch 100 --learning_rate 1e-4 --margin 1.0 --num_diffusion_iters 100

python main.py --dataset ICEWS105-15-few --data_path ./ICEWS05-15-few --few 5 --data_form Pre-Train --prefix icews05-15_frecd_iter100_lr1e-3_margin1 --device 0 --batch_size 256 --g_batch 1024 -epo 10000 --eval_epoch 100 --checkpoint_epoch 100 --learning_rate 1e-4 --margin 1.0 --num_diffusion_iters 100

python main.py --dataset ICEWS18-few --data_path ./ICEWS18-few --few 5 --data_form Pre-Train --prefix icews18_frecd_iter100_lr1e-3_margin1 --device 0 --batch_size 128 --g_batch 1024 -epo 10000 --eval_epoch 100 --checkpoint_epoch 100 --learning_rate 1e-4 --margin 1.0 --num_diffusion_iters 100

Testing

Use pre-trained checkpoints for evaluation:

python main.py --dataset XXXX --data_path ./XXXXXX --few 5 --data_form Pre-Train --device 0 --batch_size 256 --g_batch 1024 -epo 3000 --eval_epoch 100 --checkpoint_epoch 50 --learning_rate 1e-4 --margin 1.0 --num_diffusion_iters 100 --state_dir /FRECD-main/state --state_dict_filename checkpoint/state_dict_XXX.ckpt --prefix icewsXX_frecd_iter100_lr1e-3_margin1_2022_cuda:0_XXXXXXXXXX  --step dev/test

Acknowledgements

This repository is based on NP-FKGC and ReCDAP. We appreciate the efforts of the original authors and thank them for their excellent work.