Installation

CausalRL is imported as crl. Install from PyPI or source.

PyPI status

Available on PyPI (as of January 26, 2026).

Install from PyPI

python -m pip install causalrl
python -m pip install "causalrl[all]"

Install from source

git clone https://github.com/gsaco/causalrl
cd causalrl
python -m pip install -e .

Optional extras

Choose extras when you need them (use causalrl[...] for PyPI installs or .[...] for editable installs):

python -m pip install "causalrl[docs]"
python -m pip install "causalrl[benchmarks]"
python -m pip install "causalrl[notebooks]"
python -m pip install "causalrl[behavior]"
python -m pip install "causalrl[d4rl]"
python -m pip install "causalrl[rlu]"

Recommended extras

• Use .[behavior] if you need to estimate behavior propensities.
• Use .[notebooks] to run the walkthrough notebooks locally.
• Use .[docs] if you plan to build the documentation site.

Common pitfalls

• Behavior policy estimation requires scikit-learn.
• FQE and some estimators require torch.
• evaluate currently assumes discrete action spaces.
• The walkthrough notebooks call behavior estimation; install .[notebooks] (includes scikit-learn) or .[behavior] before running them.

Workflow schematic

Typical CausalRL workflow: data -> estimand -> estimator -> diagnostics -> sensitivity -> report.

Sanity check

python - <<'PY'
import crl
print(crl.__version__)
PY