Overview

The dagsampler package provides a configurable causal data generator.

Main features:

• custom and random DAG generation

• Continuous, binary, and categorical variables

• Structural forms: linear, polynomial, interaction, sigmoid, cos, sin, stratum_means

• Optional element-wise post_transform (tanh, sin, cos, exp_neg_abs, sqrt_abs, relu, sign)

• Noise models: additive (gaussian, student_t, gamma, exponential, laplace, cauchy, uniform), multiplicative, and heteroskedastic

• Cross-type mechanisms: continuous → categorical (threshold) and categorical → continuous (stratum_means, with optional metric parents)

• Random structural weight controls:

  • random_weight_low

  • random_weight_high

  • random_weight_min_abs (excludes near-zero coefficients)

• force_uniform_marginals for balanced exogenous binary / categorical draws

• Template helpers for chain, fork, collider, and independence configurations

• Reproducible sampling with separate structure/data seeds

• Optional d-separation CI oracle output