Implement of LGraphormerHTC: Label-guided Graphormer for Hierarchy Text Classification
Python >= 3.6
torch >= 1.6.0
transformers >= 4.11.0
datasets
torch-geometric == 1.7.2
torch-scatter == 2.0.8
torch-sparse == 0.6.12
Preprocess
Please download the original dataset and then use these scripts.
The original dataset can be acquired in the repository of HDLTex. Preprocessing code could refer to the repository of HiAGM and we provide a copy of preprocessing code here. Please save the Excel data file Data.xlsx in WebOfScience/Meta-data as Data.txt.
cd data/WebOfScience
python preprocess_wos.py
python data_wos.py
python keywords_cooccurrence.py
The original dataset can be acquired here. Place the unzipped folder nyt_corpus inside data/nyt (or unzip nyt_corpus_LDC2008T19.tgz inside data/nyt).
cd data/nyt
python data_nyt.py
python keywords_cooccurrence.py
The preprocessing code could refer to the repository of reuters_loader and we provide a copy here. The original dataset can be acquired here by signing an agreement. Place rcv1.tar.xz and lyrl2004_tokens_train.dat (can be downloaded here) inside data/rcv1.
cd data/rcv1
python preprocess_rcv1.py ./
python data_rcv1.py
python keywords_cooccurrence.py
Train
usage: train.py [-h] [--lr LR] [--data DATA] [--batch BATCH] [--early-stop EARLY_STOP] [--device DEVICE] [----output_dir OUTPUT_IDR] --name NAME [--update UPDATE] [--model MODEL] [--wandb] [--arch ARCH] [--layer LAYER] [--graph GRAPH] [--prompt-loss]
[--low-res] [--seed SEED]
optional arguments:
-h, --help show this help message and exit
--lr LR Start learning rate. Default: 3e-5.
--data {WebOfScience,nyt,rcv1} Dataset
--batch BATCH Batch size
--early-stop EARLY_STOP Epoch before early stop
--device DEVICE cuda or cpu. Default: cuda
--output_dir OUTPUT_DIR the dir to save the checkpoint
--name NAME A name for different runs
--update UPDATE Gradient accumulate steps
--wandb Use wandb for logging
--seed SEED Random seed
Checkpoints are in OUTPUT_DIR/checkpoints/DATA-NAME. Two checkpoints are kept based on macro-F1 and micro-F1 respectively (checkpoint_best_macro.pt, checkpoint_best_micro.pt)
Example:
python train.py --name test --batch 16 --data nyt