zinc-30/aveqa

the reproduction code of the paper AVEQA

AVEQA

the reproduction code of the paper "Learning to Extract Attribute Value from Product via Question Answering: A Multi-task Approach"

Usage

Use the following command to run the train.py or test.py

python train.py --device=cuda

The model would run in 1 GPU in defualt, to change the setting, change the number in find_gpus in train.py and test.py

Put the aepub dataset in the ./dataset folder as an txt file (./dataset/publish_data.txt), and the processed dataset would be save to './dataset/aePub'

Use the ae_pub.py to generate and preprocess the dataset.

Modify the config.json to set the parameter and the dataset path.

The training.py would automatically go through the training and testing pipeline and generate the training and testing dataset

After training, the training accuracy would be stored in the training_metric.json at the root directory.

Requirment

Python 3.7

scikit-learn

torch 1.11.0

transformer 4.18.0

You may find other packages in the requirement.txt and requirement_conda.txt for pip and conda environment.

aveqa's Issues

Dockerfile

If you update the data/weight/model, I will make a dockerfile for easy reproduction. Thanks!