Fit the CPSS-guided exact-null submodel endpoint

The selected core is refitted from scratch using an economy-SVD Moore–Penrose operator. All coefficients outside the core are set exactly to zero. The function never copies full-model coefficients and never forms an OLS Gram inverse.

Usage

fit_cpss_sm(X, y, core_set = NULL, selection = NULL, rank_tolerance = 1e-10)

Arguments

X

Numeric analysis-sample design matrix.

y

Numeric analysis-sample response.

core_set

One-based selected core indices, or a cpss_core_selection object.

selection

Alternatively, an object returned by cpss_select_core(). Named features are aligned to X by identity.

rank_tolerance

Relative economy-SVD rank tolerance.

Value

A cpss_sm_fit object with original-scale coefficients and exact-null diagnostics.

References

Belloni, A. and Chernozhukov, V. (2013). Least squares after model selection in high-dimensional sparse models. Bernoulli, 19, 521–547. doi:10.3150/11-BEJ410