Jie Liu, Yuanhao Huang, Ritambhara Singh, Jean-Philippe Vert, William Stafford Noble
Source code for the MMD-MA algorithm is available here.
Following are tables, in tab-delimited text and numpy formats, containing all data necessary to reproduce the experiments reported in the paper. All of the data may be downloaded together in this file.
| Simulation 1 | Simulation 2 | Simulation 3 | methyl-expr | |
| Domain 1 input vectors | 300 x 2 txt | 300 x 2 txt | 300 x 3 txt | |
| Domain 1 mapping | 2 x 1000 txt | 2 x 1000 txt | 3 x 1000 txt | |
| Domain 1 mapped vectors | 300 x 1000 txt | 300 x 1000 txt | 300 x 1000 txt | 61 x 2486 txt |
| Domain 1 kernel | 300 x 300 npy | 300 x 300 npy | 300 x 300 npy | 61 x 61 npy |
| Domain 2 input vectors | 300 x 2 txt | 300 x 2 txt | 300 x 3 txt | |
| Domain 2 mapping | 2 x 2000 txt | 2 x 2000 txt | 3 x 2000 txt | |
| Domain 2 mapped vectors | 300 x 2000 txt | 300 x 2000 txt | 300 x 2000 txt | 61 x 2486 txt |
| Domain 2 kernel | 300 x 300 npy | 300 x 300 npy | 300 x 300 npy | 61 x 61 npy |
| Domain 1 MMD-MA output | 300 x 5 npy | 300 x 5 npy | 300 x 5 npy | 61 x 5 npy |
| Domain 2 MMD-MA output | 300 x 5 npy | 300 x 5 npy | 300 x 5 npy | 61 x 5 npy |
Please note that, to ensure comparability of kernels between the two data modalities, each kernel was linearly rescaled to have a maximum value of 100 (divide the matrix by its maximum value and then multiply the result by 100).
Please email questions to william-noble@uw.edu and drjieliu@umich.edu.