Improve tutorials for experimental partners

[justin13601]

No description provided.

[justin13601]

Use MIMIC Demo data for integrated tutorials with MEDS, ACES, modelling etc.

[Oufattole]

People expressed interest in how to use ACES on MIMICIV. Is the plan to add a MIMICIV tutorial with some explanations of what aces is doing under the hood? Does it make sense for that to be in this repo or in the MEDS-DEV one? I currently have a tutorial in meds-torch that has a step where we pull some aces tasks. I copied it below, maybe you could use it as a start (you probably should use the MEDS-DEV tasks though):

Use ACES to extract labels using a task config definition:

We can manually extract the supervised task labels from our meds dataset using aces. First install aces:

conda create -n aces python=3.12
conda activate aces
pip install es-aces==0.5.0
pip install hydra-joblib-launcher

Second, run the following command to extract the supervised task labels:

TASKS=(
    "mortality/in_hospital/first_24h"
    "mortality/in_icu/first_24h"
    "mortality/post_hospital_discharge/1y"
    "readmission/30d"
)
for TASK_NAME in "${TASKS[@]}"; do
    SINGLE_TASK_DIR="${MIMICIV_MEDS_DIR}/tasks/${TASK_NAME}"
    mkdir -p $SINGLE_TASK_DIR # create a directory for the task
    cp MIMICIV_INDUCTIVE_EXPERIMENTS/configs/tasks/${TASK_NAME}.yaml "${SINGLE_TASK_DIR}.yaml"
    aces-cli --multirun hydra/launcher=joblib data=sharded data.standard=meds data.root="$MIMICIV_MEDS_DIR/data" "data.shard=$(expand_shards $MIMICIV_MEDS_DIR/data)" cohort_dir="$TASKS_DIR" cohort_name="$TASK_NAME"
done

[justin13601]

We considered a MIMIC-IV tutorial, but was not sure about how to effectively do so (explain what ACES is doing under the hood) without display snippets of the data. Do you have any ideas? If we don't display any data, wouldn't it be just calling a series of general commands/python functions, which may not be as helpful as we want a end-to-end tutorial to be...

[Oufattole]

Ahhh good point, you can't really show much under the hood stuff due to privacy constraints. How about demonstrating

1. a MIMICIV a workflow of getting cohort stats on various cohorts on MIMICIV cohorts as aggregate statistics are safe (number of patients, prevalence of binary prediction target, and other capabilities of interest)
2. how file structures work, specifically giving a visual of what the input and output directories should look like in your sharded vs unshareded workflows on MIMICIV. I'm not sure if this is the best way to demonstrate file structures, but I took a stab at it for meds-tab in the usage guide.