crabbymetrics

Rust-backed econometrics with NumPy arrays and a Python fit / predict / summary interface. This site documents v0.9.0.

Install and fit

pip install --upgrade crabbymetrics==0.9.0

import numpy as np
import crabbymetrics as cm

rng = np.random.default_rng(42)
x = rng.normal(size=(400, 2))
y = 0.5 + x @ np.array([1.0, -0.5]) + rng.normal(size=400)
model = cm.OLS()
model.fit(x, y)
result = model.summary(vcov="hc1")
print("Coefficients:", result["coef"])
print("Robust SE:", result["coef_se"])
print("Predictions:", model.predict(x[:3]))

Coefficients: [ 0.99250001 -0.5722471 ]
Robust SE: [0.05152025 0.0458896 ]
Predictions: [ 1.3856094   0.69463719 -0.7031842 ]

Pass finite NumPy arrays with the documented dtype and shape. Fits mutate the estimator in place; they do not return a fitted object. Summaries are dictionaries. The package’s only Python runtime dependency is NumPy.

Find an estimator

Task	Reference	Worked example
Linear regression and robust inference	OLS, fixed effects, TwoSLS	OLS, IV
Shrinkage and nonlinear prediction	Ridge, ElasticNet, bagged polynomials	Ridge tuning, feature pipelines
Binary, multiclass, and count outcomes	Logit, MultinomialLogit, Poisson	Prediction contracts
Event timing and recurrent events	CoxPH, AndersenGill, WeibullPH	Survival models
Treatment effects and balancing	AIPW, DML, BalancingWeights, MPE_CBPS	Double ML, Chronos LTV
Panel counterfactuals	SyntheticDID, MatrixCompletion, AugmentedBalancing	Staggered adoption
Dynamic treatment regimes	RegressionBlip, ParallelTrendsSNMM, DynamicCovariateBalance	Timing contracts and simulations
Custom moments or objectives	GMM, MEstimator, Optimizers	GMM, custom Poisson

The API overview lists all public classes, functions, and live signatures. The v0.9 migration guide covers changed numerical and inference behavior. A dense machine-readable guide is available in llms.txt.

Simulations and case studies

• Variance estimators: Monte Carlo coverage under heteroskedasticity and overdispersion.
• Semiparametric estimators: misspecification, cross-fitting, and balancing.
• Panel DGP comparisons and real panel cases.
• Randomized sketching: approximation error and computational tradeoffs, not a coverage guarantee.
• Estimator scaling: bounded timing runs with explicit versions, failures, and reference implementations.
• A First Course in Causal Inference: executable chapter examples using Peng Ding’s replication data.

Implementation and reproduction

The OLS binding walkthrough, native Poisson solver, and MEstimator callback bridge explain how Python and Rust divide the work. See Rebuilding the docs for dependencies, data provenance, fresh simulation execution, and deployment checks.