Please, Cease Employing Static Hyper-Parameters in Active Learning: A Comprehensive Framework Comparison

Acquisition Functions Visualization

Random Sampling Baseline

Coreset

Entropy

BALD

Installation

conda create -n alrm python=3.9 -y
conda activate alrm  
git clone https://github.com/Nearpit/ALRM.git
cd ALRM
pip install .

To add a new acquisition function

1. Add a new class to the acquisition folder inhereting some of the base classes.
2. Add the relative name for the --a argument.
3. Import the added class to the __init__.py file in the acquisition folder.

To add a new dataset

1. Add a new class to the dataset folder inhereting some of the base classes.
2. Import the added class to the __init__.py file in the acquisition folder.
3. Initialize json file in the dataset/configs folder (see example.json)
   1. n_features - the number of features of the added dataset
   2. n_instances - the number of instances of the added dataset
   3. n_classes - the number of classes of the target space
   4. batch_size - training batch size
   5. budget - the amount of AL iterations (0 < x < train_size - n_labeled)
   6. metrics_dict - metrics of the classifier