statsmodels.robust.covariance.CovDetMCD.fit#

CovDetMCD.fit(h, *, h_start=None, mean_func=None, scale_func=None, maxiter=100, options_start=None, reweight=True, trim_frac=0.975, maxiter_step=100)[source]#

Compute minimum covariance determinant estimate of mean and covariance

Parameters:

hint: Number of observations in evaluation set for minimizing determinant.
h_startint: Number of observations used in starting mean and covariance.
mean_func, scale_funccallable or None: Mean and scale function for initial standardization. Current defaults, if they are None, are median and mad, but default scale_func will likely change.
maxiterint: Maximum number of iterations for the c-step of the best candidate solution.
options_startNone or dict: Options for the starting estimators. Currently not used. TODO: which options? e.g., for OGK
reweightbool: If reweight is true, then a reweighted estimator is returned. The reweighting is based on a chisquare trimming of Mahalanobis distances. The raw results are in the results_raw attribute.
trim_fracfloat in (0, 1): Trim fraction used if reweight is true. Used to compute quantile of chisquare distribution with tail probability 1 - trim_frac.
maxiter_stepint: Number of iteration in the c-step. In the current implementation a small maxiter in the c-step does not find the optimal solution.

Returns:

CovDetMCDResult: Named tuple with mean, cov, method and extra attributes depending on reweight. See CovDetMCDResult for details.