Code to reproduce the numerical experiments in the paper.

./notebook/run_simple.ipynb is a simple notebook example to run implemented algorithms.

Package info:

./src/mdata: data generation and preprocessing

./src/mmodel: pytorch models

./src/method: DA algorithms

Each experiment is conducted using 10 distinct seeds. Hyperparameters are chosen via a grid search based on validation accuracy, where the same set of hyperparameters is adopted across seeds. Refer to Appendix D.3 and D.4 for details.

The following commands reproduce experiments in Section 5.1.1, 5.2.1, 5.3, and 5.5.

# exp = SCM_1, SCM_2, SCM_3, SCM_4 
#       MNIST_rotation_1, MNIST_rotation_2, MNIST_rotation_3, MNIST_rotation_4 
#       CelebA_1, CelebA_2, CelebA_3, CelebA_4, CelebA_5, CelebA_6
#       DomainNet
# seed = 1, 2, ..., 10

python run_exp.py --seeds seed --exp exp

The following commands reproduce experiments in Section 5.4.

# seed = 1, 2, ..., 10
# algorithm = CIP, 
#             DIP-Pool, JointDIP-Pool                                   (using covariates of unlabeled target data)
#             DIP-Pool_target_labeled, JointDIP-Pool_target_labeled     (using covariates of labeled target data)

python camelyon.py --seeds seed --algorithm algorithm --download False

The following commands reproduce experiments in Section 5.1.2 and 5.2.2.

# DIP on SCM III (Figure 3, 4)
# alpha = 0., .25, .5, .75
python scm3_dip.py --alpha alpha 

# DIP on MNIST III (Figure 3, 6)
# alpha = 0., .25, .5, .75, .9, .95
python mnist3_dip.py --alpha alpha

# JointDIP on SCM III (Figure 3) 
python scm3_jointdip.py 

# JointDIP on MNIST III (Figure 3) 
python mnist3_jointdip.py 

All data should be put in the ./data folder.

• SCM, MNIST: automatically generated / downloaded by python run_exp.py.

• CelebA: https://mmlab.ie.cuhk.edu.hk/projects/CelebA.html.

  • Extract data from img_align_celeba.zip to folder data/CelebA/img_align_celeba.

  • Move list_attr_celeba.txt to folder data/CelabA/attr.

• Camelyon17: https://github.com/p-lambda/wilds.

  • python camelyon.py with --download flag set to True or False.

• DomainNet: https://ai.bu.edu/M3SDA/.

  • Extract data from the zip files (clipart.zip, infograph.zip, painting.zip, quickdraw.zip, real.zip, sketch.zip) into their respective folders under data/domainNet/ (clipart, infograph, painting, quickdraw, real, sketch).

  • Place the text files (clipart_train.txt, clipart_test.txt, infograph_train.txt, infograph_test.txt, etc.) in the data/domainNet directory.