Intermediary data is not stored on this repo and has to be generated. Run the following files if this is your first time:
checks.ipynbto create folder structure (where the following scripts will save their data)Eligibility Criteria.ipynbto generate list of eligible patient idsGenerating Dataset Pipeline.ipynbto generate a json dataset of the admissions dataSplit dataset.ipynbto generate the train-validation-test splits.Register best model.ipynbto create, train and save a model's weights and hyperparametersCreate Variational Outputs.ipynbto use the previous model (or any other really) to generate several forward passes on the validation set
Having done this the analysis notebooks can work:
Accuracy as a function of confidence.ipynbfor the general uncertainty analysis- TODO
To be able to run and reproduce this repo, create a conda environment:
conda env create -n <env_name> -f environment.yml