uoo723 / M2m_text

Implementation of "M2m: Imbalanced Classification via Major-to-minor Translation" for text data.

M2m_text

Prerequisites

• python==3.7.9
• torch==1.7.1
• numpy==1.19.2
• tqdm==4.54.1
• click==7.1.2
• logzero==1.6.3
• scikit-learn==0.24.1
• ruamel.yaml==0.16.12
• gdown==3.12.2
• pandas==1.1.5
• nltk==3.5
• gensim==3.8.3
• scipy==1.5.2
• transformers==4.2.1
• mlflow==1.15.0

$ pip install -r requirements.txt

Docker

If you prefer to use docker, follow below instructions.

Prerequisites

• cuda >= 11.0
• Docker >= 19.03.5
• nvidia-docker

Build Image

$ docker build -t [image_name]

Or use our pre-built image realappsdev/m2m-text

Run docker cotainer

$ docker run --rm -t -d --init --name [container_name] --gpus all -v $PWD:/workspace -w /workspace --ipc=host --net=host [image_name]
$ docker exec -it [cotainer_name] zsh  # Enter shell

Run

$ ./scripts/run_model.sh

MLflow Integration

You can track experiments with mlflow.

Run MLflow UI

$ ./scripts/mlflow_ui.sh  # Enter http://[your-ip]:5000