Mismatch in compute_feature_label_alignment

[martijnsiepel01]

Hi!

I am trying to write a custom labeler for the task of predicting Hospital Acquired Infections (HAI, infections that occur 48+ hours after admission). What I have done previously is creating a custom labelers for different SNOMED codes and run the erhshot benchmark code using femr 0.0.20, which worked fine. I used a simple custom implementation of the FirstDiagnosisTimeHorizonCodeLabeler.

However, for a HAI, I want to change the prediction time to 48 hours after admission by adding time to the admission time(similar to how the x-ray task moves the prediction task to some time before the result gets registered). I attempted this in two ways, using both an implementation of a Labeler labeler, as well as an implementation of a WithinVisitLabeler.

Both labelers work from a technical point of view, they run and produce legit looking label files. But, once I get to to the evaluation portion of the script (7_eval), I run into the following issue:

---

Traceback (most recent call last):
File "/mnt/data/inbox/Martijn/Scripts Martijn/ehrshot/ehrshot-benchmark/ehrshot/bash_scripts/../7_eval.py", line 212, in
patient_ids, label_values, label_times, feature_matrixes = get_labels_and_features(labeled_patients, PATH_TO_FEATURES_DIR)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/mnt/data/inbox/Martijn/Scripts Martijn/ehrshot/ehrshot-benchmark/ehrshot/utils.py", line 337, in get_labels_and_features
join_indices = compute_feature_label_alignment(label_patient_ids,
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/mnt/data/inbox/Martijn/Scripts Martijn/ehrshot/ehrshot-benchmark/ehrshot/utils.py", line 283, in compute_feature_label_alignment
raise RuntimeError(f"Could not find match for {label_pids[i]} {label_dates[i]}, closest is {feature_pids[j]} {feature_dates[j]}")
RuntimeError: Could not find match for 115967095 1223596560000000, closest is 115967095 1223580540000000

Apparantly there is a mismatch between my labels and features. How can I solve this? Does my labeler have an issue?

I would appreciatie some thoughts!
HAI_Labeler.txt
HAI_WithinVisitLabeler.txt

Best