Commands to reproduce Figure 6 (Colored MNIST experiment):
python PseudoLabel_MNIST.py --train_S True --train_T True --cor_prob 0.95
python PseudoLabel_MNIST.py --train_S True --train_T True --cor_prob 0.97

Commands to reproduce Figure 7 (Interpolation between pseudolabeling and entropy minimization):
python PseudoLabel_MNIST.py --train_S True --pseudolabel True --num_rounds 3
python PseudoLabel_MNIST.py --pseudolabel True --num_rounds 6
python PseudoLabel_MNIST.py --pseudolabel True --num_rounds 30

Commands to reproduce Figure 8 (Gaussian mixture experiment):
python3 PseudoLabel_Gaussian.py entropy_min
python3 PseudoLabel_Gaussian.py pseudolabel

Commands for CelebA experiment:
python process_data.py
python PseudoLabel_celebA.py --target_w 0 --n_epoch 300
python PseudoLabel_celebA.py --resume --target_w 1 --n_epoch 400

Computational Resource:
For CelebA, 1 NVIDIA GPU: Source training 20min, source + target training 22min.
All other experiments were run on Macbook Pro and computation time was negligible.
