Appendix for " Handling Missing Data with Graph Representation Learning "

For GAIN, we use the source code released by the authors. Here we report the running clock time for feature imputation of different methods at test time. We adapt the same setting as in Section 4.1 and the results are shown in Appendix C. G The Douban dataset has 3000 observations and 3000 features. The Y ahooMusic dataset has 1357 observations and 1363 features. Inductive matrix completion based on graph neural networks.