physionet_sepsis_challenge_2019 | Physionet 2019 Early Detection

Once the repo has been cloned locally, setup a python environment with ``python==3.6`` and run ``pip install -r requirements.txt``. you must download the official PhysioNet training data. This consists of two folders of .psv files. These folders should be placed in data/raw/training_{A, B}. To setup the data run ``src/data/make_dataframe.py``. Recommended since the dataset is very large, for offline testing and experimentation, a test environement was created by placing a /test/ folder within data. The structure is then /data/test/{external, interim, processed, raw} that mimics the structure in /data/{external, interim, processed, raw}. This can be setup by running ``src/data/test_data.py`` after the data setup step. This will take a handful of septic cases, mix with some non-septic cases and generates save files in ``data/test`` that will be loaded if running in the test environemnt. To generate a model run ``src/models/experiments/main.py`` and input a name for the experiment, this will end up being the name of the folder in ``models/`` that will contain run information. Inside this file is a dictionary of options that can be edited in model training. It includes choices of what features to compute and any associated hyperparameters. For example changing: ~ { …​ featurecolumns: [SBP, DBP, HR], featureorder: [5, 6], …​ } ~ will run two experiments, the first computing signatures of Systolic BP, Diastolic BP and HR to order 5 the second to order 6, with other options as specified. Once run, this will save model metrics and the probabilities output for each timestamp in ``models/{test, if in test env}/experiment_name/run_num/``. Here experiment_name is the input specified on running ``src/models/experiments/main.py`` and ``run_num`` is a number that increases by 1 for every option combination. In the example above the order 5 computation is run number 1 and the order 6 is run number 2. To use the full dataset rather than experiment in the test environment the paths in ``definitions.py`` must be changed: ~ # From # Setup paths DATA_DIR = ROOT_DIR + /data/test MODELS_DIR = ROOT_DIR + /models/test.
DATA_DIR = ROOT_DIR + /data MODELS_DIR = ROOT_DIR + /models ~ I had a simple if statement that chose between depending on whether I was working locally or via ssh.