_fits

cultivars.engine._internals._fits = <module 'cultivars.engine._internals._fits' from '/home/runner/work/cultivars/cultivars/src/cultivars/engine/_internals/_fits.py'>

the raw numeric output of an estimator, before result assembly.

Every class here is a frozen, slotted dataclass and nothing else. A fit carries what an estimator produced – coefficients, residuals, a likelihood, a parameter count – together with the quantities derivable from those alone. It carries no estimation logic: how a fit is produced belongs to the model that owns the specification, in cultivars._internals._models.

That separation is what keeps this module a leaf. It imports numpy, the dataclass machinery, and one predictor protocol; nothing else in the package depends on it in the other direction, so a fit record can be constructed, compared, or serialized without dragging in an optimizer.

The hierarchies are shallow and share fields only where the fields genuinely mean the same thing across the group. _ConditionalVarianceFit is the one multi-level case: its two subclasses parameterize the variance path in ways that are not comparable, so only the path itself and the two quantities derived from it live on the shared base.

Type:

Fit records