checkmodel

One-call diagnostics for statistical models in Python, ported from R's performance package.

pip install checkmodel

from checkmodel import check_model

report = check_model(fit)   # a statsmodels, scikit-learn, PyMC or bambi model
report.plot()

check_model() works out what kind of model it was given and runs the checks that make sense for it. A linear regression gets normality, homoscedasticity, multicollinearity and influential observations, plus autocorrelation for time series. A logistic regression gets binned residuals, a count model overdispersion and zero-inflation, and a mixed model singularity and random-effect normality. Every model starts with a posterior predictive check.

Where to go next

• Getting started: install, check a model, read the results, compare models.
• Examples: worked notebooks for linear, logistic, mixed and Bayesian models, and a guide for R users.
• Supported models: which check works on which model.
• API reference: every function and result type.
• Comparison with R: function-by-function mapping and the few deliberate differences.

Tested against R

Every statistic is compared with R (performance 0.18.2, lme4, lmtest, car, MASS, rstanarm, loo and posterior) on 27 reference models, to a tolerance of 1e-4. The Bayesian statistics are checked on the same posterior draws as R.