Downloadable from this Google Drive link
Each dataset has splits:
- train : to train the model
- dev : to tune hyper parameters
- test : for final test that gives numbers and figures in the paper
- (test1) : for future test
- censor.conf : to specify missingness mechanism details (see section 5 in paper for more technical details)