StanfordShahLab

/

clmbr-t-base

like 82

Follow

Stanford Shah Lab 87

There's seems to be a few compatibility issue when trying to use this model to fine-tune on MEDS data. The current preprocessor is incompatible with meds_reader.SubjectDatabas, FEMR==0.2.3 is incompatible with femr.datasets.PatientDatabase, which means the only way to load MEDS data is using transformer.Dataset. But, when I use transformer.Dataset, the data does not load in the appropriate format for "index = femr.index.PatientIndex(dataset, num_proc=4)", since the latest MEDS format splits events in different rows rather than having one row per patient.

Has anyone been able to configure a fine-tuning pipeline with MEDS 5.3?