explain

explain

Core explain() entry point: turn a fitted model into an LLM explanation.

Functions

Name	Description
explain	Explain a statistical model’s output using an LLM.
suggest_code	Suggest code snippets to run next based on a model explanation.

explain

explain.explain(
    model_object,
    client,
    context=None,
    language=None,
    audience='novice',
    verbosity='moderate',
    style='markdown',
    prompt_dir=None,
)

Explain a statistical model’s output using an LLM.

Supported Models

This function supports fitted model objects from the following packages: * statsmodels: - OLS (Ordinary Least Squares regression) - GLM (Generalized Linear Models, e.g., Binomial, Gamma, NegBinomial) - MixedLM (Linear Mixed Effects models) - ARIMA / SARIMAX (Time Series models) - PHReg (Cox Proportional Hazards regression) * lifelines (optional): - CoxPHFitter (Cox Proportional Hazards survival models) - WeibullAFTFitter / LogNormalAFTFitter / LogLogisticAFTFitter (Parametric Accelerated Failure Time survival models) * pygam (optional): - GAM / LinearGAM / LogisticGAM / PoissonGAM (Generalized Additive Models)

Parameters

Name	Type	Description	Default
model_object	Any	A fitted statistical model object from a supported library (e.g., a results object from statsmodels).	required
client	chatlas.Chat	A chatlas Chat client (e.g. from chatlas.ChatOpenAI() or chatlas.ChatAnthropic()). Never mutated by this function.	required
context	str	Additional context about the data or research question to provide to the LLM.	None
language	str	The language the explanation should be written in (e.g. “Spanish”, “French”, “Mandarin Chinese”). If None (the default), no language constraint is added and the LLM will typically respond in the same language as the input/context or its default language.	None
audience	str	The target audience: one of “novice” (default), “student”, “researcher”, “manager”, or “domain_expert”.	'novice'
verbosity	str	The desired level of detail: one of “brief”, “moderate” (default), or “detailed”.	'moderate'
style	str	The output format style: one of “markdown” (default), “html”, “json”, “text”, or “latex”.	'markdown'
prompt_dir	str	Custom prompts directory path to override package prompts.	None

Returns

Name	Type	Description
	dict	A dictionary with keys: text, model_type, audience, verbosity, style.

suggest_code

explain.suggest_code(explanation)

Suggest code snippets to run next based on a model explanation.

Parameters

Name	Type	Description	Default
explanation	dict	The dictionary returned by explain().	required

Returns

Name	Type	Description
	str	A formatted string of suggested Python diagnostics code.