Simulated data

dml_rs.datasets (Rust: dml_core::datasets) provides the data generating processes of DoubleML's make_* functions, for examples, tests and simulation studies. The functions keep DoubleML's names, parameters and defaults.

PythonRust

from dml_rs.datasets import make_plr_CCDDHNR2018, make_did_CS2021

x, y, d = make_plr_CCDDHNR2018(n_obs=500, dim_x=20, alpha=0.5, random_state=42)
dml_rs.PLR(ml_l, ml_m).fit(x, y, d).summary

df = make_did_CS2021(n_obs=1000, random_state=42)    # long-format panel
x_cols = ["Z1", "Z2", "Z3", "Z4"]
dml_rs.StaggeredDID(ml_g, ml_m).fit_frame(df, "y", "id", "t", "d", x_cols)

use dml_core::datasets::{DidCs2021, PlrCcddhnr2018};

let sim = PlrCcddhnr2018 { n_obs: 500, ..Default::default() }.generate(42)?;
let fit = Plr::new(Ols::new(), Ols::new()).fit(&sim.to_data()?)?;

let panel = DidCs2021::default().generate(42)?;
let fit = StaggeredDid::new(Ols::new(), LogisticRegression::default()).fit(&panel.to_panel()?)?;

The data are drawn in Rust from a seeded generator. random_state makes them reproducible; without it the seed is drawn from NumPy's global generator, so np.random.seed works as well. The draws are not identical to DoubleML's, but the distributions are the same. The tests compare moments and output formats with DoubleML's generators on large samples.

return_type is "array" (a tuple of arrays, the default) or "DataFrame" (pandas, with DoubleML's column names). DoubleML's data classes have no counterpart, since dml-rs models take arrays. Generators that DoubleML returns as a dict (with oracle values) or a DataFrame return the same here.

Function	Design	Returns
make_plr_CCDDHNR2018	partially linear regression (Chernozhukov et al., 2018)	x, y, d
make_plr_turrell2018	partially linear regression with a random covariance (Turrell, 2018)	x, y, d
make_confounded_plr_data	PLR with an unobserved confounder of given strength	dict with oracle values
make_pliv_CHS2015	partially linear IV with many instruments	x, y, d, z
make_pliv_multiway_cluster_CKMS2021	PLIV with two-way clustering	x, y, d, clusters, z
make_lplr_LZZ2020	logistic partially linear model, binary outcome	x, y, d, p
make_plpr_CP2025	static panel with correlated unit effects	DataFrame
make_irm_data	binary treatment	x, y, d
make_confounded_irm_data	binary treatment with an unobserved confounder	dict with oracle values
make_irm_data_discrete_treatments	multi-valued treatment	dict with oracle values
make_iivm_data	binary instrument (LATE)	x, y, d, z
make_ssm_data	sample selection, missing at random or nonignorable	x, y, d, z, s
make_heterogeneous_data	heterogeneous treatment effects	dict with data, effects and effect function
make_did_SZ2020	two-period DiD (Sant'Anna and Zhao, 2020), panel or cross-sections	z, y, d, t
make_did_CS2021, make_did_cs_CS2021	staggered adoption (Callaway and Sant'Anna, 2021)	DataFrame
make_simple_rdd_data	regression discontinuity	dict

In Rust, each design is a struct with public fields (DoubleML's parameters, with Default) and a generate(seed) method. The results convert to the library's data types: Simulated::to_data, SimulatedPanel::to_static_panel, SimulatedStaggered::to_panel / to_cross_sections and SimulatedRdd::to_data.