This folder contains a suite of standalone Python simulations and a profiling utility that demonstrate how RobustiPy can be used on synthetic data to estimate effects, perform cross-validated model selection, and visualize robustness, as well as to profile runtime performance.

Each script runs from the command line (no notebook required) and saves figures to a ../figures directory.

• sim1_vanilla.py
  Basic OLS robustness demo on simulated data.

  • Data: n=100 observations; 1 main regressor x1 + 4 controls z1…z4.
  • Outcome: Linear in x1 and controls with Gaussian noise.
  • Model: OLSRobust with bootstrapping (draws=1000) and 10-fold CV.
  • Output: SVG plots and summary table.

• sim2_longitudinal.py
  Panel-style simulation with group-level heterogeneity.

  • Data: 1,000 groups × 10 obs each; correlated covariates from a factor model + shocks; group-specific coefficients.
  • Model: OLSRobust with group='group', rescale_y=True, rescale_z=True.
  • Output: HQIC-based selection plots (PDF) and summary table.

• sim3_constants.py
  Multiple focal regressors with fixed coefficients.

  • Data: n=1000; x1 and z1 as focal regressors, z2…z7 as controls; correlated multivariate normal design.
  • Model: OLSRobust with draws=1000, 10-fold CV.
  • Output: HQIC plots (PDF) and summary.

• sim4_binary.py
  Binary outcome example using logistic-style robustness.

  • Data: n=10,000; covariates from a factor-structured covariance; latent index thresholded at its median to form y1 ∈ {0,1}.
  • Model: LRobust with x=['x1'], controls z1…z7, draws=1000, 10-fold CV.
  • Output: HQIC plots for several specs (PDF) and summary.

• sim5_multipley.py
  Multiple dependent variables in a robust OLS framework.

  • Data: n=1000; 5 covariates (x1…x5); 4 outcomes (y1…y4) generated from stacked β-vectors; 4 controls (z1…z4).
  • Model: OLSRobust with multiple outcomes, draws=1000, 10-fold CV, and rescaling of y/x/z.
  • Output: Plots (two-panel layout) showing robustness across specs, plus summary.

• time_profiler.py
  Performance benchmarking tool for RobustiPy.

  • Purpose: Tests runtime under different numbers of controls, draws, and CV folds for both OLSRobust and LRobust.
  • Features:
    • Skips already-completed runs by tracking them in CSV files.
    • Runs both OLS and logistic variants with the same synthetic data generation.
    • Varies control set size (CONTROL_VARS subsets), draws (logarithmic sequence), and folds (FOLDS_LIST).
    • Handles segmentation faults gracefully by restarting the process.
  • Output: Figures (PDF) and CSV logs of runtime per configuration.

Install RobustiPy and basic scientific Python packages:

pip install robustipy numpy pandas matplotlib