sig_mp: Estimate of the variance from Gavish and Donoho (2014) based...
In dcgerard/hose: Higher-order spectral estimators

Description Usage Arguments Details Value Author(s) References

Returns a normalized squared median singular value that is a consistent estimate of the variance under a specific asymptotic regime.

1	sig_mp(dmed, N, p)

`dmed`	A positive numeric. The median singular value of the data matrix
`N`	A positive integer. The row dimension.
`p`	A positive integer. The column dimension.

Under the asymptotic regime where the rank of the mean is fixed and the signal to noise ratio is fixed, the squared median singular value divided by the variance and the larger dimension size will converge to the median of the Marcenko-Pasture distribution. Hence, the squared median singular value divided by the larger dimension size and the median of the Marcenko-Pastur distribution will converge to the variance of the data matrix. This estimator works really well when the rank is much less than the dimensions of the matrix and really poorly otherwise.

sig2_est A positive numeric. The estimate of the variance of the data matrix.

David Gerard

Gavish, M., & Donoho, D. L. (2014). The optimal hard threshold for singular values is 4/sqrt(3). Information Theory, IEEE Transactions on, 60(8), 5040-5053.

dcgerard/hose documentation built on Aug. 1, 2019, 12:11 a.m.