Examples

All examples use synthetic data and run without MIMIC-IV access. Swap in a real MimicTableLoader path to run against real data — the rest of the pipeline is identical.

---

Jupyter notebooks

01 — Getting started

A module-by-module walkthrough that demonstrates every clinops component in isolation. Good starting point if you are new to the library.

Topics covered:

Section	Module	Key class
Schema validation	clinops.ingest	FlatFileLoader, ClinicalSchema
Outlier clipping	clinops.preprocess	ClinicalOutlierClipper
Unit normalization	clinops.preprocess	UnitNormalizer
ICD harmonization	clinops.preprocess	ICDMapper
Temporal windowing	clinops.temporal	TemporalWindower
Lag/rolling features	clinops.temporal	LagFeatureBuilder
Imputation	clinops.temporal	Imputer
Splitting	clinops.split	PatientSplitter, TemporalSplitter, StratifiedPatientSplitter

---

02 — ICU mortality prediction pipeline

An end-to-end pipeline that takes raw (synthetic) MIMIC-IV-style tables and produces ML-ready X_train / y_train / X_test / y_test arrays for a 48-hour ICU mortality prediction task.

Pipeline stages:

raw EHR data
  → cohort definition  (ICU stays with ≥48h data)
  → ingest & validate  (FlatFileLoader + ClinicalSchema)
  → clip outliers      (ClinicalOutlierClipper)
  → ICD harmonization  (ICDMapper → chapter-level comorbidity flags)
  → align to admission (CohortAligner → first 48h only)
  → temporal windows   (TemporalWindower, 8h window / 4h stride)
  → patient split      (StratifiedPatientSplitter)
  → imputation         (Imputer fitted on train only)
  → lag features       (LagFeatureBuilder)
  → merge comorbidities
  → scikit-learn ready arrays

---

Script example

e2e_example.py

A single runnable Python script that exercises the full v0.1 pipeline. Useful for quick smoke-testing after installation.

pip install clinops
python examples/e2e_example.py

---

Running the notebooks locally

pip install clinops jupyter
jupyter notebook examples/notebooks/

Or with JupyterLab:

pip install clinops jupyterlab
jupyter lab examples/notebooks/

The notebooks install no additional dependencies beyond clinops itself. The optional scikit-learn baseline in notebook 02 requires pip install scikit-learn.

---

Adapting to real MIMIC-IV data

Replace the synthetic data generation block in either notebook with:

from clinops.ingest import MimicTableLoader

tbl      = MimicTableLoader("/data/mimic-iv-2.2")
cohort   = [10000032, 10000980, ...]   # your subject_id list

charts   = tbl.chartevents(subject_ids=cohort)
icustays = tbl.icustays(subject_ids=cohort, with_los_band=True)
adm      = tbl.admissions(subject_ids=cohort)
dx       = tbl.diagnoses_icd(subject_ids=cohort)

Everything from the preprocessing step onwards is identical.