Clarification on readm_ts.normalizer in paths.py: Is it just a renamed ihm_ts2.0.normalizer?

[glory6-CYL]

Hi authors,

Thank you for sharing the excellent codebase for CMCM! I am currently working on a rigorous reproduction of your experiments on both the MIMIC-III and MIMIC-IV datasets.

While reviewing the data preprocessing pipeline and configuration, I noticed a detail in paths.py regarding the normalizers. Both the Readmission and In-Hospital Mortality (IHM) tasks for both datasets are pointed to the exact same file, readm_ts.normalizer:

MIMIC4_READM_NORMALIZER_PATH = ROOT_PATH / "data_mimic4/readm_ts.normalizer"
MIMIC4_IHM_NORMALIZER_PATH = ROOT_PATH / "data_mimic4/readm_ts.normalizer"
MIMIC3_READM_NORMALIZER_PATH = ROOT_PATH / "data_mimic3/readm_ts.normalizer"
MIMIC3_IHM_NORMALIZER_PATH = ROOT_PATH / "data_mimic3/readm_ts.normalizer"

As we know, the standard mimic3benchmark and your extended mimic4extract pipelines generate standard normalizers like ihm_ts2.0.input_str:previous.start_time:zero.normalizer, but they do not generate a readm_ts.normalizer by default.

Given that your paper specifies sampling "every two hours" for both tasks, I wanted to clarify:

Is readm_ts.normalizer actually just the standard ihm_ts2.0.normalizer file that was simply renamed locally for convenience? - If YES: I will just point all these paths to my generated ihm_ts2.0...normalizer to proceed with the training.

If NO (i.e., it was specifically fitted on the Readmission patient subset): Could you please kindly provide the readm_ts.normalizer file, or share the script snippet used to generate it, so I can align my data distribution exactly with your reported results?

Thank you in advance for your time and clarification!