EntLM
The source codes for EntLM.
Dependencies:
Cuda 10.1, python 3.6.5
To install the required packages by following commands:
conda create --name EntLM python=3.6.5
conda install pytorch torchvision torchaudio pytorch-cuda=11.8 -c pytorch -c nvidia
$ pip install -r requirements.txt
To download the pretrained bert-base-cased model:
$ cd pretrained/bert-base-cased/
$ sh download_bert.sh
Few-shot Experiment
Run the few-shot experiments on CoNLL 5-shot with:
sh scripts/run_conll.sh
By default, this runs 4 rounds of experiments for each of the sampled datasets. You can also run 10/20/50-shot experiments by editing the line FILE_PATH=dataset/conll/5shot/ in scripts/run_conll.sh .
Label word selection
You can run the label word selection process by:
sh scripts/count_freq.sh
This will build a label_map file such as dataset/conll/label_map_timesup_ratio0.6_multitoken_top6.json in the dataset path.
You can try different method by changing "--sort_method" to ["LM", "data", "timesup"].
Or you can try different ratio/virtual_number by changing "--filter_ratio" and "--top_k_num".