cleverly#

Causal studies with TMLE · Alpha

Estimate. Validate. Stress-test.

Build a causal study on TMLE. Name the question, inspect the identification argument, fit, and then probe the result with validation, diagnostics, and sensitivity analyses.

It is named for TMLE’s clever covariate. Even the jokes in this toolbox are targeted.

Get started

Browse the Python API

Warning

cleverly is alpha software. Unsupported combinations fail before fitting. The public API may still change, so pin a version for reproducible work. Read the identification statement your analysis returns.

Install from PyPI#

python -m pip install cleverly

Python 3.11 or newer is required. The installation guide covers optional backends, development environments, and reproducible version-pinned installs.

One workflow, explicit decisions#

  1. DesignDeclare the observed data
  2. EstimandName the causal question
  3. IdentifyInspect assumptions
  4. EstimateConfigure and fit
  5. ResultAssess and report

The separation is deliberate. Changing an estimation method does not silently change the scientific question. Follow the analysis workflow from formulation through diagnostics and reporting.

Choose your path#

Getting started

Install the package, fit one average treatment effect, and learn how to read the result.

Getting started
Workflow

Move from a causal question and observed-data design to diagnostics and reporting.

Workflow
User guide

Choose estimands, interventions, learners, methods, inference, and assessments.

User guide
Technical reference

Trace every implementation to theory, citations, source, provenance, and evidence.

Technical reference
Examples

Follow complete point-treatment, intervention, longitudinal, and assessment workflows.

Examples
Python API

Look up supported objects, signatures, attributes, methods, and return types.

Python API

What is implemented, and what is planned#

The implementation matrix lists every implementation family the package ships. Each row names the theory, the source module, the external provenance, and the correctness evidence.

The roadmap gives source-backed work in one binding grid. A second grid keeps future investigations out of the implementation sequence until a paper supplies the missing theory. A roadmap entry is an accepted direction and not a release claim.