Installation

ehr2rl supports Python 3.10 and later. The base install is enough for synthetic data, rewards, and behavior-policy estimation. Real-data and export features live behind optional extras so that a plain install stays small.

Base install

pip install ehr2rl

This installs NumPy, pandas, scikit-learn, and SciPy. It is all you need for the synthetic quickstart.

Optional extras

Extra	Install	Adds	Needed for
d3rlpy	pip install "ehr2rl[d3rlpy]"	d3rlpy	to_d3rlpy export
bigquery	pip install "ehr2rl[bigquery]"	google-cloud-bigquery, google-cloud-bigquery-storage, db-dtypes, pyarrow	MIMIC-IV v3.1 through BigQuery
all	pip install "ehr2rl[all]"	Both of the above	The full BigQuery-to-d3rlpy workflow

Quote the argument ("ehr2rl[d3rlpy]") so shells such as zsh do not treat the square brackets as a glob pattern.

Importing ehr2rl never requires an extra. Optional packages are imported only when you call a function that needs them, and that function raises an ImportError naming the extra to install.

Contributor installs

From a clone of the repository:

Extra	Install	Use
dev	pip install -e ".[dev]"	Tests, Ruff, and mypy
dev,all	pip install -e ".[dev,all]"	Tests that exercise the BigQuery and d3rlpy paths
docs	pip install -e ".[docs]"	Building this documentation site (Python 3.12 or later)

Data access

ehr2rl does not ship, mirror, or provide access to MIMIC-IV. To work with real data you need credentialed access through PhysioNet and must follow its data use agreement. Everything else in these guides works without it.