CausalData at a glance

CausalData is the light-weight input container used across Causalis. It wraps a pandas DataFrame and records which columns are the outcome, the treatment, and the confounders.

Quick start

Note: Internally, the stored DataFrame is trimmed to only these columns: [outcome, treatment, confounders].

API essentials

• Init parameters

  • df: pandas DataFrame (no NaNs)
  • treatment: name of the treatment column (numeric)
  • outcome: name of the outcome column (numeric)
  • confounders: one or more confounder column names (numeric)

• Properties

  • outcome: pandas Series
  • treatment: pandas Series
  • confounders: list[str] of confounder column names

• Method

  • get_df(columns=None, include_treatment=True, include_outcome=True, include_confounders=True) -> DataFrame Selects columns by name and/or by role. Returns a copy.

Validation (on construction)

1. No missing values anywhere in df.
2. All referenced columns must exist.
3. Outcome, treatment, and confounders must be numeric (int/float).
4. None of these columns can be constant (zero variance).
5. Any two used columns having identical values is disallowed (raises ValueError).
6. Duplicate rows across the used columns trigger a warning (not an error).

Common snippets

Tips

• For categorical confounders, encode them numerically (e.g., one-hot) before creating CausalData.
• If you see the duplicate-rows warning, consider deduplicating if duplicates are unintended.
• repr shows the stored shape and declared roles for quick inspection.