statsmodels.robust.resistant_linear_model.RLMDetSMM.fit#

RLMDetSMM.fit(h=None, scale_binding=False, start=None)[source]#

Estimate the model

Parameters:

hint or None, optional: The size of the initial sets for the S-estimator. If None (default), max(nobs // 2 + 1, k_params + 1) is used, the same rule as CovDetS.fit’s h_start default. Unused if start is provided.
scale_bindingbool, optional: If true, then the scale is fixed in the second stage M-estimation, i.e., this is the MM-estimator. If false, then the high breakdown point M-scale is used also in the second stage M-estimation if that estimated scale is smaller than the scale of the preliminary, first stage S-estimator.
starttuple or None, optional: If None, then the starting parameters and scale for the second stage M-estimation are taken from the first stage S-estimator. Alternatively, the starting parameters and starting scale can be provided by the user as tuple (start_params, start_scale). In this case the first stage S-estimation is skipped.

Returns:

statsmodels.robust.robust_linear_model.RLMResults: Results instance

Notes

maxiter and other optimization parameters are still missing (todo).

If scale_binding is false, then the estimator is a standard MM-estimator with fixed scale in the second stage M-estimation. If scale_binding is true, then the estimator will try to find an estimate with lower M-scale using the same scale-norm rho as in the first stage S-estimator. If the estimated scale, is not smaller than then the scale estimated in the first stage S-estimator, then the fixed scale MM-estimator is returned.