Closed QZJ-2003 closed 12 hours ago
Have you used the train/val/test generator script from this repo? The only post-processing (that can exclude patients) done to the *.csv files generated by the MIMIC-sepsis extraction script is skipping of trajectories with length=1 (see here). The number reported in the paper is len(train_dict) + len(val_dict) + len(test_dict) = 18,585
.
The 2024 ICLR paper mentions that 'We focus our analysis on patients that fulfill Sepsi-3 criteria (Singer et al., 2016), 18,585 in total'. However, I use MIMIC-III v1.4 to reproduce the data process of this code, and then the result number of
.csv
file is 18,918. Why do I have more data? How can I check the reason? And How can I verify the correctness of my result?