FewShotText
This repository contains code for the paper A Neural Few-Shot Text Classification Reality Check
Environment setup
# Create environment
python3 -m virtualenv .venv --python=python3.6
# Install environment
.venv/bin/pip install -r requirements.txt
# Activate environment
source .venv/bin/activate
Fine-tuning BERT on the MLM task
model_name=bert-base-cased
block_size=256
dataset=OOS
output_dir=transformer_models/${dataset}/fine-tuned
python scripts_transformers/run_language_modeling.py \
--model_name_or_path ${model_name} \
--output_dir ${output_dir} \
--mlm \
--do_train \
--train_data_file data/${dataset}/full/full-train.txt \
--do_eval \
--eval_data_file data/${dataset}/full/full-test.txt \
--overwrite_output_dir \
--evaluate_during_training \
--logging_steps=1000 \
--line_by_line \
--logging_dir ${output_dir} \
--block_size ${block_size} \
--save_steps=1000 \
--num_train_epochs 20 \
--save_total_limit 20 \
--seed 42
Training a few-shot model
To run the paper's experiments, simply use the utils/scripts/runner.sh file.
Reference
If you use the data or codes in this repository, please cite our paper:
@article{dopierre2021neural,
title={A Neural Few-Shot Text Classification Reality Check},
author={Dopierre, Thomas and Gravier, Christophe and Logerais, Wilfried},
journal={arXiv preprint arXiv:2101.12073},
year={2021}
}
92 Jan 7, 2023
23 Dec 21, 2022
319 Dec 30, 2022
34 Dec 25, 2022
399 Jan 8, 2023
82 Dec 24, 2022
26 Nov 30, 2022
116 Dec 19, 2022
603 Jan 5, 2023
246 Dec 19, 2022
224 Dec 28, 2022
94 Dec 12, 2022
36 Nov 22, 2022
124 Jan 4, 2023
91 Dec 23, 2022
34 Oct 8, 2022
835 Jan 8, 2023
238 Jan 4, 2023