CLI Reference ¶

OneEHR exposes six top-level commands:

preprocess — bin features, split patients
train — train ML/DL models
test — evaluate on test set
analyze — cross-system comparison and analysis
plot — render publication-quality figures
convert — convert raw datasets to OneEHR format

View the live interface with:

oneehr --help

`oneehr preprocess`¶

oneehr preprocess --config <toml>

Bins dynamic events, generates labels, splits patients, and writes the run manifest. This is the required first step for every run.

Flag	Required	Description
`--config`	Yes	Path to TOML experiment config

Outputs are written to {output.root}/{output.run_name}/preprocess/.

`oneehr train`¶

oneehr train --config <toml> [--force]

Trains all models defined in [[models]] and writes checkpoints under {run_dir}/train/{model_name}/.

Flag	Required	Description
`--config`	Yes	Path to TOML experiment config
`--force`	No	Overwrite existing train directory

`oneehr test`¶

oneehr test --config <toml> [--force]

Runs all trained models and configured [[systems]] on the held-out test split. Writes predictions.parquet and metrics.json under {run_dir}/test/.

Flag	Required	Description
`--config`	Yes	Path to TOML experiment config
`--force`	No	Overwrite existing test directory

`oneehr analyze`¶

oneehr analyze --config <toml> [--module NAME]

Reads test/predictions.parquet and writes structured analysis outputs under {run_dir}/analyze/.

Flag	Required	Description
`--config`	Yes	Path to TOML experiment config
`--module`	No	Run a single analysis module instead of all

Available modules:

comparison — cross-system metrics comparison with bootstrap CI
feature_importance — native importance for tree models, SHAP, permutation importance
fairness — demographic parity, equalized odds, predictive parity, SMD
calibration — temperature scaling, isotonic regression, ECE
statistical_tests — DeLong, McNemar, BH FDR correction
missing_data — missingness analysis per feature

`oneehr plot`¶

oneehr plot --config <toml> [--figure NAME ...] [--style STYLE] [--output DIR]

Renders publication-quality figures from test/analyze results.

Flag	Required	Description
`--config`	Yes	Path to TOML experiment config
`--figure`	No	Specific figure(s) to render (default: all available)
`--style`	No	Journal style preset: `default`, `nature`, `lancet`, `wide`
`--output`	No	Output directory for figures (default: `{run_dir}/figures/`)

Available figures: roc, pr, confusion, calibration, decision_curve, forest, fairness, training_curves, significance, missing_heatmap, cohort_flow, kaplan_meier, attribution, waterfall.

`oneehr convert`¶

oneehr convert --dataset <name> --raw-dir <path> --output-dir <path> [--task TASK]

Converts a raw clinical dataset into OneEHR's three-table format (dynamic.csv, static.csv, label.csv).

Flag	Required	Description
`--dataset`	Yes	Source dataset: `mimic3`, `mimic4`, or `eicu`
`--raw-dir`	Yes	Path to raw dataset directory
`--output-dir`	Yes	Output directory for converted CSVs
`--task`	No	Label task to export (default: all tasks as separate files)

Supported datasets and tasks ¶

Dataset	Expected layout	Available tasks
`mimic3`	Flat directory with `ADMISSIONS.csv`, `PATIENTS.csv`, `LABEVENTS.csv`, etc.	`mortality`, `readmission`, `los_3day`, `los_7day`
`mimic4`	`hosp/` and `icu/` subdirectories	`mortality`, `readmission`, `los_3day`, `los_7day`
`eicu`	Flat directory with `patient.csv`, `lab.csv`, etc.	`mortality`, `los_3day`, `los_7day`

Example:

# Convert MIMIC-III with mortality labels
oneehr convert --dataset mimic3 --raw-dir ~/data/mimic-iii/ --output-dir data/mimic3/ --task mortality

# Convert all available labels
oneehr convert --dataset mimic4 --raw-dir ~/data/mimic-iv/ --output-dir data/mimic4/

← Reference Configuration →