ARC-TIGERS

Testing Imbalanced cateGory classifiERS

Installation

git clone https://github.com/alan-turing-institute/ARC-TIGERS
cd ARC-TIGERS
python -m pip install .

Usage

### scripts/dataset_download.py This script downloads a reddit dataset and saves in an appropriate place in the parent directory. It takes as arguments:

• dataset_name: the name of the dataset to load from huggingface, for example bit0/reddit_dataset_12
• target_subreddits: a list subreddits being used for the experiment, this should be in .json format.
• max_rows: The maximum number of rows to use in the resultant dataset, this should be an integer. It saves a .json file called filtered_rows containing the data in a subdirectory named using the dataset name and the maximum number of rows.

### scripts/dataset_generation.py This script generates train and test splits from the downloaded reddit dataset(s). It currently takes as arguments:

• data_dir: the path to the dataset being used to form the splits
• split: the specific subreddit split to generate. Defined in the dictionary DATASET_COMBINATIONS in /data/utils.
• r: The ratio of target subreddits to non-target subreddits

The script saves two csv files, train.csv and test.csv within a subdirectory splits, these contain the train and evaluation splits and are of roughly equal size.

Contributing

See CONTRIBUTING.md for instructions on how to contribute.

License

Distributed under the terms of the MIT license.