v1.4.1

enhanced cache_util to remove the data dir associated with the cachin…

v1.4.0: Update runner util (#98)

* upgraded cehrbert_data dependency

* removed max_position_embeddings from computing generate_prepared_ds_path

v1.3.9: Enhance meds to OMOP logic for handling the problem list records (#95)

* updated the logic for constructing the visit type for meds input

* changed a few attributes to properties in patient_block

* added logic for disassociating the problem list records from the visit because they could occur years before

* updated the logic to infer whether the visit should have a discharge facility code associated with it

* provide a list of features to the data generator to convert meds to cehrbert patients

* use the smaller integers to represent the meds generated data

* removed unused import

* updated hf_dataset_mapping.py to work with event generator

* updated hf_dataset_mapping for backward compatibility

* updated the type of visit_end_datetime

* set all timestamps to timestamp[us] when converting meds to cehrbert patient

* fixed a bug where the inferred_visit_type() was called at the wrong place

* added a function in patient_block.py to merge the patient block that is subsumed by another

* fixed the merging visit logging information

* disassociate records from the visits if they occur outside the visit

* added newline to the debug_log in merge_patient_blocks

* added a default none value to visit_complete_event_time

* added E-I visit type update in patient block merging

* removed a newline char

* swap the patient blocks if the the longer visit is an outpatient visit and the shorter visit (being subsumed) is an inpatient or emergency visit

* added one more check on the end attribute in the event

* only extract the end datetime of the record corresponding to the visit table

* remove the visit_type and discharge_facility events in patient block merging

* updated patient_block_merge test

* manually clean up the cached files generated from Dataset.from_generator when meds format is used

* collect additional cached files from the downstream datasets derived from the generator based Dataset

* use CacheFileCollector to remove all the cached files generated from the dataset mapping functions

* moved CacheFileCollector to a separate python module

* updated the logic for identifiying the visit start and visit end when using the meds data

* updated cache_util to handle Dataset and DatasetDict accordingly

* removed the logic for bounding the event time stamps

* updated cahce file collection logic for fine-tuning

* fixed type checking in add_cache_files

v1.3.8

only generate patients who have visits in _meds_to_cehrbert_generator…

v1.3.7: Cehrbert streaming for meds (#90)

* fixed the streaming option for cehrbert pretraining when streaming is enabled for the meds data

* fixed a potential bug where the CehrbertConfig was misconfigured

* lossened up dependencies

* disabled parallism when streaming is enabled

* added the missing meds_reader sample data

* removed pydant

v1.3.6

lossened up the huggingface dependencies (#89)

v1.3.5

changed to the partial match to the meds_birth code instead of the ex…

v1.3.4: New cehrbert data backward compatible (#86)

* added patient_splits_folder to evaluation

* support the new cehrbert_data

* added backward compatibility support for the old spark cehrbert_data application

* fixed any none values in concept_values

* removed the renaming logic

* added try catch in case the compute_metrics in finetuning fails

v1.3.3

upgraded cehrbert_data to 0.0.5 (#84)

v1.3.2

added patient_splits_folder to evaluation (#83)