XLearner<T>

Models & Types · AiDotNet.CausalInference

Implements the X-Learner (Cross-learner) for treatment effect estimation.

Inheritance: CausalModelBase<T> → XLearner<T>

Implements: IGeneratedParameterRegistrar<T>

For Beginners

X-Learner is a sophisticated meta-learner that adapts to the data by using cross-fitting. It’s especially effective when treatment and control groups have different sizes, as it leverages information from both groups more efficiently.

How It Works

How it works (5 stages):

  • Train μ₀(X) and μ₁(X) using T-Learner approach
  • Impute treatment effects: D₁ᵢ = Y₁ᵢ - μ₀(X₁ᵢ) for treated, D₀ᵢ = μ₁(X₀ᵢ) - Y₀ᵢ for control
  • Train τ₁(X) on D₁ (treated imputed effects) and τ₀(X) on D₀ (control imputed effects)
  • Estimate propensity score e(X) = P(T=1|X)
  • Combine: τ(X) = e(X)·τ₀(X) + (1-e(X))·τ₁(X)

Key Insight: The weighted combination uses propensity scores to give more weight to the model trained on the larger group, making X-Learner robust to imbalanced data.

Pros and Cons:

  • Pro: Excellent for imbalanced treatment groups
  • Pro: Can outperform T-Learner when one group is much smaller
  • Pro: Adapts to the data structure through propensity weighting
  • Con: More complex, requires fitting 5 models
  • Con: Propensity estimation can be sensitive

When to use:

  • When treatment/control groups are imbalanced
  • When you want state-of-the-art CATE estimation
  • When you have sufficient data for multiple model fitting

Reference: Künzel et al., “Metalearners for Estimating Heterogeneous Treatment Effects” (2019)

Example

var covariates = new Matrix<double>(new double[,]
{
    { 45, 1 }, { 52, 0 }, { 38, 1 }, { 61, 0 }, { 47, 1 }, { 55, 0 }
});
var treatment = new Vector<int>(new int[] { 0, 1, 0, 1, 0, 1 });   // who was treated
var outcome = new Vector<double>(new double[] { 3.1, 7.2, 2.8, 8.0, 3.4, 7.6 });
            
var result = new AiModelBuilder<double, Matrix<double>, Vector<double>>()
    .ConfigureModel(new XLearner<double>(lambda: 0.01))
    .Build(covariates, treatment, outcome);
            
// One effect per person, over the covariates alone.
Vector<double> cate = result.EstimateTreatmentEffect(covariates);

Constructors

Constructor Summary
XLearner(double lambda) Creates a new X-Learner.
XLearner(int maxIterations) Creates a new X-Learner.

Properties

Property Summary
double Lambda { get; } Gets the L2 regularization strength.

Methods

Method Summary
IFullModel<T, Matrix<T>, Vector<T>> WithParameters(Vector<T> parameters)
ValueTuple<T, T> EstimateATE(Matrix<T> x, Vector<int> treatment, Vector<T> outcome)
ValueTuple<T, T> EstimateATT(Matrix<T> x, Vector<int> treatment, Vector<T> outcome)
Vector<T> EstimateCATEPerIndividual(Matrix<T> x, Vector<int> treatment, Vector<T> outcome)
Vector<T> EstimateTreatmentEffect(Matrix<T> features) Estimates the Conditional Average Treatment Effect (CATE).
Vector<T> Predict(Matrix<T> input) Standard prediction - returns treatment effects.
Vector<T> PredictControl(Matrix<T> features) Predicts outcome under control.
Vector<T> PredictTreated(Matrix<T> features) Predicts outcome under treatment.
Vector<T> PredictTreatmentEffect(Matrix<T> x)
void Fit(Matrix<T> features, Vector<T> treatment, Vector<T> outcome) Fits the X-Learner model.