README.md

Installation

Accessing the MIMIC-III data

The MIMIC-III dataset contains sensitive patient information, and as such, a training and validation procedure is needed in order to access the dataset. This can be carried out here. We use the same data preprocessing as the CAML paper, so after obtaining access to the data you can run the following notebook.

Processing the MIMIC-III data

Follow the notebook caml/notebooks/dataproc_mimic_III.ipynb.

Setting environment variables

Within the CAML folder there is a caml/constants.py file. You will need to point those variables that are currently set to CHANGEME. Use absolute paths to point to datasets and point MODEL_DIR to the experiments folder, which we create next.

Dependencies

pip install -r requirements.txt

There may be additional python dependencies you need to install, but we updated the caml code to be compatible with the more recent torch version we installed for spmlbl.

Running experiments

export PYTHONPATH="$PWD/caml"
mkdir -p experiments
mkdir -p logs
./run.sh

Verifying models

After running the command below, each model folder will be populated with two analysis files: test-analysis.json and dev-analysis.json.

./eval.sh