MIMIC Dataset for Batch RL [Feature Request]

[braj29]

Motivation

Want to add the MIMIC Dataset, generally used for Sepsis Treatment for benchmarking Batch-RL algorithms in a easy & efficient way. This dataset is used in many BatchRL papers for sepsis treatment. See: https://github.com/asjad99/MIMIC_RL_COACH

Solution

Use D4RL wrapper (https://github.com/Farama-Foundation/D4RL) to download and load the dataset ready for training using simple python commands

Alternatives

For larger datasets it maybe best to download the files partially and then delete them as training progressing while simultaneously downloading remaining data

Additional context

Checklist

• [x] Had initial discussion with @vmoens
• [ ] Check if it is allowed to download and use the dataset directly, while verifying or redirecting users who have fulfilled the requirements to physionet for permission. Currently it says, to download MIMIC, you must become a credentialed user on PhysioNet
• [ ] Create a wrapper to download, verify/redrect and load the dataset

[vmoens]

The D4RL dataset format is a bit messy IMO. Can we interact with the dataset directly without recurring to D4RL?