mmcdermott/comprehensive_mtl_ehr

Source code for a comprehensive analysis of MTL over EHR timeseries data.

comprehensive_mtl_ehr's Issues

Data preprocessing pipeline

Thank you for releasing the source code.
Will you provide the extraction code for the eICU extraction system and the code for splitting each dataset? Also, is there any other parts of the data preprocessing pipeline that haven't been released, except for MIMIC-Extract?

Suggested use of pre-processed data

Hi @mmcdermott,

really awesome project and resource! I was wondering if there's any suggested use of the pre-processed data and splits. I see that some of the .pkl files are actually dictionaries. Do you recommend pandas to pre-process them into a pytorch dataloader or something else? Sorry I cannot seem to find any relevant code in the repo.

Would you like to share the code of eICU extraction system

Dear author,

Thanks very much for this meaningful work! I found in the paper (A Comprehensive EHR Timeseries Pre-training Benchmark) that you mentioned “Extraction code for our eICU extraction system will be released publicly after publication.” However, I haven’t found the code yet. Would you like to share that part of the code? Thanks!

nans in icd labels

Hi,

I noticed in your data for a lot of patients the icd labels are all nan. How did you deal with this when training for the icd prediction task? Did you set the label for these patients to "Unknown"?
It looks like it in the code, but I wonder, because in your paper in Table 7 the majority class accuracy for Unknown is label 0 with 100%.

Thank you and best regards

Import Errors in v2_push

Blocking #1. Reported by @hunterlang

"For example,

comprehensive_MTL_EHR/latent_patient_trajectories/representation_learner/dataset.py

Line 14 in 611d833

from ..data_utils import convert_notes_to_features_eff, convert_notes_to_features_bret, reformat_notes, prepare_continuous_labels, tokenize_notes

dataset.py imports reformat_notes from data_utils, but I don’t see that function defined there (and this gave an import error running Scripts/run_model.py)

The other one I found was evaluator.py importing SelfAttentionTimeSeries from adapted_model, which also doesn’t seem to be defined. Once I fixed those two, I was able to at least run Scripts/run_model.py.
"