This code is used for the active leraning in multi-class logistic regression. The active learning methods included are: random, kmeans, entropy, varratios, BAIT, and FIRAL (our method). The entry point of the code is run.py. An example of conductin an experiment could be the following command:

python run.py --data='cifar10' --method='FIRAL' --b=10 

which will use FIRAL to select 10 points to label from the unlabeled pool of CIFAR10 dataset. More parameter settings can be found in run.py.

In folder "data/", we have features after pre-processing (via unsupervised learning approaches) for three datasets: MNIST, CIFAR10, ImageNet50. File "imagenet50_classes.txt" contains names of the 50 classes that are randomly selected from the original ImageNet dataset.

