isSingular

Detect whether a fitted model is at or near the boundary of the parameter space — i.e., whether any variance component has collapsed to zero. Mirrors lme4::isSingular().

isSingular(result, tol=0.0001)[source]

Return True if the model is at or near the boundary of the parameter space.

A model is singular when one or more variance components have collapsed to zero — the corresponding diagonal entry of the relative covariance factor Lambda_theta is less than tol. This matches the behaviour of lme4::isSingular().

Parameters:

  • result – A fitted CrossedLMEResult.

  • tol – Tolerance for declaring a diagonal entry “effectively zero”. Defaults to 1e-4, matching lme4.

Return type:

bool

Returns:

bool – True if any variance component is at the boundary.

Parameters:

Examples

>>> interlace.isSingular(result)
False

Example

import interlace

result = interlace.fit("rt ~ condition", data=df, groups=["subject", "item"])

if interlace.isSingular(result):
    print("Warning: one or more variance components are at the boundary.")
    print(result.boundary_flags)  # {'subject': False, 'item': True}

Notes

fit() automatically issues a ConvergenceWarning when isSingular() returns True. You can silence the warning if you expect a boundary fit (e.g., in simulations) and check it explicitly with this function instead.

See also