API Reference¶
Welcome to the DoubleML PySpark API documentation. This section provides comprehensive reference material for all classes and methods available in the DoubleML PySpark library.
Overview¶
DoubleML PySpark is designed around a modular architecture that enables scalable causal inference using Double Machine Learning techniques. The API follows a consistent design pattern across all estimators, making it easy to switch between different causal models.
Architecture¶
The library is built on a hierarchical class structure:
DoubleMLBase (Abstract Base Class)
└── DoubleMLScalar (Single Parameter estimation)
├── DoubleMLPLR (Partially Linear Regression)
└── DoubleMLIRM (Interactive Regression Model)
Core Classes¶
Base Infrastructure¶
| Class | Purpose | Documentation |
|---|---|---|
| DoubleMLBase | Abstract base class providing core functionality for all DoubleML estimators | Core methods, cross-fitting, inference |
| DoubleMLScalar | Single parameter estimation | Simplified estimation workflow |
Causal Models¶
| Class | Purpose | Documentation |
|---|---|---|
| DoubleMLPLR | Partially Linear Regression model | Estimates treatment effects in partially linear models |
| DoubleMLIRM | Interactive Regression Model | Estimates average treatment effects with binary treatment |
Effect Heterogeneity Tools¶
| Class | Purpose | Documentation |
|---|---|---|
| DoubleMLBLP | Best Linear Predictor | Projects CATEs and GATEs onto linear combinations of covariates |
Tuning¶
Hyperparameter Tuning API: Optuna-based hyperparameter optimization for nuisance models
| Component | Type | Description |
|---|---|---|
| DMLOptunaResult | dataclass |
Container for tuning results: best estimator, parameters, score, and Optuna study history |
| _dml_tune_optuna | function |
Performs Optuna hyperparameter tuning for a given estimator using PySpark DataFrames |