metaMIMIC

analysis of hyperparameter transferability for tabular data using MIMIC-IV database

What is this?

This is a repository containing code used to reproduce the results presented in the metaMIMIC study available here.

What is needed to reproduce the results?

To reproduce the results you will need the following:

• access to the MIMIC-IV data. Relevant information is available here: https://physionet.org/content/mimiciv/1.0/,
• a PostgreSQL database containing MIMIC-IV data. Relevant scripts are available here: https://github.com/MIT-LCP/mimic-iv/tree/master/buildmimic/postgres,
• a Python environment containing packages listed in the requirements.txt file (we suggest using the Miniconda distribution available here: https://docs.conda.io/en/latest/miniconda.html and running the conda create --name metaMIMIC --file requirements.txt command),
• code available in this repository.

It is important to mention that both the PostgreSQL database and the proposed calculations may be considered resource-heavy. It took us several days of CPU time using 48 cores and 256 GB of RAM to generate all the results.

How to reproduce the results?

Results of the experiments are already available in the respective experiment directories, but you can also reproduce them by making sure you have all the above available and following these steps:

1. Create the 1_metaMIMIC_data/metaMIMIC.csv file containing the base data for all models. To do this, provide PostgreSQL database credentials and location (username, password, host address, and database name) in the 1_metaMIMIC_data/connection_info.txt file and run the 1_metaMIMIC_data/metaMIMIC_data.py script (e.g. cd 1_metaMIMIC_data && nano connection_info.txt && python metaMIMIC_data.py on Linux).
2. Run the first two experiments. To do this, remove or rename the already provided (2-3)_metaMIMIC_experiment_(1-2)/results.csv results file (the script will not run if it is present) and run the (2-3)_metaMIMIC_experiment_(1-2)/metaMIMIC_experiment_(1-2).py script (e.g. cd 2_metaMIMIC_experiment_1 && mv results.csv original_results.csv && python metaMIMIC_experiment_1 on Linux).
3. Prepare the data needed for the third experiment. To do this, run the 4_metaMIMIC_columns/metaMIMIC_columns.py script (e.g. cd 4_metaMIMIC_columns && python metaMIMIC_columns.py on Linux).
4. Run the third experiment. To do this, remove or rename the already provided 5_metaMIMIC_experiment_3/results.csv results file (the script will not run if it is present) and run the 5_metaMIMIC_experiment_3/metaMIMIC_experiment_3.py script (e.g. cd 5_metaMIMIC_experiment_3 && ls results* | xargs -I {} mv {} official_{} && python metaMIMIC_experiment_3.py on Linux).
5. Run the last experiment. To do this, remove or rename the already provided 6_metaMIMIC_experiment_bayes/results.csv results file (the script will not run if it is present) and run the 6_metaMIMIC_experiment_bayes/metaMIMIC_experiment_bayes.py script (e.g. cd 6_metaMIMIC_experiment_bayes && mv results.csv original_results.csv && python metaMIMIC_experiment_bayes.py on Linux).

All the exemplary commands are supposed to be run from the toplevel directory of this repository.

Additional information

Directory mementoML_results contains the results from the mementoML study available here: https://arxiv.org/abs/2008.13162. Our results were prepared using the same hyperparameter grid and we used the mementoML results for a comparison.

The code we used to analyse the results and prepare figures for the metaMIMIC paper can be found in the results_analysis directory.