Multi-source Heterogeneous Domain Adaptation with Conditional Weighting Adversarial Network

• python 3.6
• tensorflow-gpu 1.4
• CUDA 8.0
• cudnn 6.0
• numpy
• scipy
• matplotlib
• scikit_learn

$ conda create -n cwan python=3.6
$ conda activate cwan

$ pip install tensorflow-gpu==1.4
$ conda install cudatoolkit=8.0
$ conda install cudnn=6.0
$ conda install scipy
$ conda install matplotlib
$ conda install scikit-learn
$ conda install matplotlib

You can download the example dataset from here (Password: 8i7c), and put in the folder of datasets.

All of the datasets can be downloaded from here (Password: 4y1q).

1. You can run this code by inputing:

python -W ignore main.py

The results should be close to 59.67 (A (D_{4096}), D (R_{2048}) -> W (S_{800})). Note that different environmental outputs may be different.

2. You can use your datasets by replacing:

source_exp = [ad.SAD, ad.SDR]
target_exp = [ad.TWS]

3. You can tune the parameters, i.e., lr_1 (learning rate for g(\dot), f(\dot)), lr_2 (learning rate for d(\dot)), T, d, beta, tau, for different applications.

4. The default parameters are: lr_1 = 0.004, lr_2 = 0.001, T = 500, d = 256, beta = 0.03, tau = 0.004.

If you find this helpful, please cite:

@article{DBLP:journals/corr/abs-2008-02714,
  author    = {Yuan Yao and
               Xutao Li and
               Yu Zhang and
               Yunming Ye},
  title     = {Multi-source Heterogeneous Domain Adaptation with Conditional Weighting
               Adversarial Network},
  journal   = {CoRR},
  volume    = {abs/2008.02714},
  year      = {2020},
  url       = {https://arxiv.org/abs/2008.02714},
  archivePrefix = {arXiv},
}