Open alisoncallahan opened 3 years ago
Update to reflect changes in the newest PR:
clmbr_create_info
--input_data_dir
clmbr_train_model
--model_dir
has been updateddiabetes_code = timelines.get_dictionary().map("ICD10CM/E11.9")
. I'm not too sure myself, @Lalaland do you have any clarification on this?convert_patient_data
is doing
2. CLMBR Pre-training.ipynb
--clmbr_create_info
--clmbr_train_model
clmbr_train_model
function, which isn't consistent with the documentation of the notebook3a. Model Training with Custom Labeler.ipynb
"Set up the labeler for the downstream task we're interested in"
the comment for the DiabetesLabeler class could use a little more detail - the description below this cell makes it sound like the labeler is labeling patient days based on whether there is a diabetes code present on that day or not. But the comment talks about a prediction task with a time horizon (also, how does one specify the time point that the horizon is relative to?).
why do timelines have dictionaries? need documentation in timeline.pyi
"this randomly selects on label per patient " should be "this randomly selects one label per patient"
for output of
ehr_ml.clmbr.featurize_patients_w_labels
(features, labels, patient_ids, day_offsets
) -- what isday_offsets
relative to? DOB of patient?"Using the trained model"
patient_indices
should beday_offsets
(based on featurize_patients_w_labels in https://github.com/som-shahlab/ehr_ml/blob/245dd3436a5dcddada41222611e8129be96cd85b/ehr_ml/clmbr/__init__.py)3b. Model Training with Patient List.ipynb