# modalities.fmri.spm.reml¶

## Functions¶

nipy.modalities.fmri.spm.reml.orth(X, tol=1e-07)

Compute orthonormal basis for the column span of X.

Rank is determined by zeroing all singular values, u, less than or equal to tol*u.max().

INPUTS:
X – n-by-p matrix
OUTPUTS:
B – n-by-rank(X) matrix with orthonormal columns spanning
the column rank of X
nipy.modalities.fmri.spm.reml.reml(sigma, components, design=None, n=1, niter=128, penalty_cov=1.2664165549094176e-14, penalty_mean=0)

ReML estimation of covariance components from sigma using design matrix.

INPUTS:

sigma – m-by-m covariance matrix components – q-by-m-by-m array of variance components

mean of sigma is modeled as a some over components[i]
design – m-by-p design matrix whose effect is to be removed for
ReML. If None, no effect removed (???)

n – degrees of freedom of sigma penalty_cov – quadratic penalty to be applied in Fisher algorithm.

If the value is a float, f, the penalty is f * identity(m). If the value is a 1d array, this is the diagonal of the penalty.
penalty_mean – mean of quadratic penalty to be applied in Fisher
algorithm. If the value is a float, f, the location is f * np.ones(m).
OUTPUTS:

C – estimated mean of sigma h – array of length q representing coefficients

of variance components

cov_h – estimated covariance matrix of h