Every model class publishes one value in its constructor as fit.options, so the options a result was produced with can be read back from the result itself. GLM and quantile fits publish the estimator-specific extensions below. The value survives store_data=False and lean=True, which it describes through retention.
Parameters
Name
Type
Description
Default
ssc
Ssc
Small-sample correction options, as built by ssc().
required
drop_singletons
bool
Whether singleton fixed-effect levels were dropped (fixef_rm="singleton").
required
drop_intercept
bool
Whether the intercept was removed from the design.
required
weights
str or None
Name of the observation-weight column, or None for an unweighted fit.
required
weights_type
('aweights', 'fweights')
Interpretation of weights: analytic or frequency weights.
"aweights"
offset
str or None
Name of the offset column of a GLM fit, or None.
required
collin_tol
float
Tolerance of the rank check that drops collinear columns.
required
solver
str
Linear solver used for the least-squares steps.
required
demeaner
MapDemeaner or LsmrDemeaner
Resolved fixed-effect demeaner configuration.
required
store_data
bool
Whether the estimation data and formula state were retained.
required
copy_data
bool
Whether the estimation data was copied before the fit, the documented exception to never mutating user input.
required
lean
bool
Whether the large fit products were dropped after estimation.
required
context
Mapping[str, Any]
Variables made available to formulaic when materializing the model matrix, as captured by capture_context.