Duarte-Almeida / BenchmarkingImbalancedLearningTechniques

Benchmarking robust model training techniques in the imbalanced setting

Setup

Generate Kaggle API Token

To fetch datasets from kaggle, we have to generate an API token into a json file. For that purpose, in the Kaggle site, go the API section in the Settings page, click "Create new token" and save the generated kaggle.json file in this directory. Then run the following commands:

sudo mkdir .kaggle
sudo cp kaggle.json .kaggle
sudo chmod 600 .kaggle/kaggle.json
sudo chown `whoami`: .kaggle/kaggle.json
export KAGGLE_CONFIG_DIR='.kaggle/'

Dependencies

To install the dependencies, just run:

pip install -r requirements.txt

Running the project

To get metrics from applying this baseline classifier, run the following command:

python3 main.py -dataset baf [-oversampling <oversampling_strategy>] [-undersampling <oversampling_strategy>] [-label_smoothing] [-plot_scores]
-data_subsampling <proportion>]

If -label_smoothing is set, the targets are smoothed and -plot_scores allows for a visualization of the class conditional distribution of the model's scores. The value determines the amount of training examples to retain from the original training set (useful for very big datasets). As of now, the following undersamping and oversampling strategies are supported

• <oversampling_strategy> can be ['SMOTE', 'ADASYN', 'KMeansSMOTE', 'RACOG']
• <undersampling_strategy> can be ['RUS', 'TomekLinks', 'ENN']

Only on the first time will the dataset be fetched from Kaggle. In subsequent executions, it will be stored in memory.