ProfBressan / ActiveLearning

Implementations

• AL Strategies: ['EN', 'MS', 'LC', 'RBE', 'DBE', 'MST-BE']
• Classifiers: ['SVM', 'k-NN', 'RF', 'NB']
• Datasets: ['LEA-53']

Installation

ActiveLearning requires Python >= 3.5

• numpy
• scipy
• scikit-learn
• tqdm
• pandas
• googledrivedownloader
• modAL

You can install directly with pip:

pip3 install git+https://github.com/ProfBressan/ActiveLearning.git

Usage

First, you need a folder named 'datasets' which, for each dataset, must contain at least 2 CSV files (features, labels) respecting the following naming convention:

• features: '<dataset_name>_features.csv' required
• labels: '<dataset_name>_labels.csv' required
• filenames: '<dataset_name>_filenames.csv' optional

You can check an example dataset under 'datasets' folder of this repository.

After that, you can run the following example (example.py):

import numpy as np
import drActiveLearning
from drActiveLearning import AL_Strategy, AL_Parameters
from drActiveLearning.classification import Classifier
from drActiveLearning.dataset import Dataset

if __name__ == '__main__':
    np.random.seed(1) #for reproducibility on some al strategies

    print('AL Strategies:', AL_Strategy.get_names())
    print('Classifiers:', Classifier.get_names())
    print('Datasets:', Dataset.get_names())

    al_params = AL_Parameters(dataset_name='LEA-53', classifier_name='NB', strategy_name='MS', max_iterations=20)
    results = drActiveLearning.run(al_params=al_params, n_splits=1)
    results.save_json('LEA-53-results')

Authors

• Daniel H. Acorsi Alves, Rafael S. bressan {dhaalves, profbressan}@gmail.com