zifengding / fitcarl

FITCARL

This is the code for the paper Improving Few-Shot Inductive Learning on Temporal Knowledge Graphs using Confidence-Augmented Reinforcement Learning (paper (appendices also included)) accepted to ECML/PKDD 2023.

Datasets are already preprocessed.

To run 3-shot meta-training on ICEWS14-OOG:

python3 main_OOG.py --data_path data/ICEWS14/processed_data --pretrain --emb_nograd --history_encoder gru --score_module att --few 3 --valid_epoch 10 --max_action_num 50 --entity_learner nn --adaptive_sample --sector --save_path logs/icews14_few3 --beam_size 100 --conf --cuda

To run 1-shot meta-training on ICEWS14-OOG:

python3 main_OOG.py --data_path data/ICEWS14/processed_data --pretrain --emb_nograd --history_encoder gru --score_module att --few 1 --valid_epoch 10 --max_action_num 50 --entity_learner nn --adaptive_sample --sector --save_path logs/icews14_few1 --beam_size 100 --conf --cuda

ICEWS18-OOG and ICEWS0515-OOG are only trainable with more than 15GB GPU memory.

You can also try as you wish. Just change the corresponding arguments from the above examples.

Code is developed based on TITer.

fitcarl's Issues

data preparation

Hello!
I am trying to run your model on my own dataset, and I'm having a hard time figuring out how you exactly generated the data files (from the main dataset files) you need to train your model. Could you please provide a script so that I can prepare the suitable data files for this repository from my own dataset?

As far as I understood, one needs to generate following files to be able to use your model:
back_metrain.txt
back_metatest.txt
stat.txt
back_metavalid.txt
background_graph_for_rl.txt