Struct RandomInterceptStats 

pub struct RandomInterceptStats {
    pub xtx: Mat<f64>,
    pub xty: Vec<f64>,
    pub yty: f64,
    pub n: usize,
    pub p: usize,
    pub group_n: Vec<f64>,
    pub group_s: Mat<f64>,
    pub group_t: Vec<f64>,
}

Sufficient statistics for a single-random-intercept fit.

Everything the REML criterion needs, reduced from the data once. See the module documentation for why this is the whole story.

Fields

xtx: Mat<f64>

X'X, p x p.

xty: Vec<f64>

X'y, length p.

yty: f64

y'y.

n: usize

Number of observations.

p: usize

Number of fixed-effect columns.

group_n: Vec<f64>

Observations per group. Empty groups are dropped on construction, so every entry is positive.

group_s: Mat<f64>

Column sums of X within each group, G x p.

group_t: Vec<f64>

Sum of y within each group, length G.

Implementations

impl RandomInterceptStats

pub fn from_design<T: BixverseFloat>( x: MatRef<'_, T>, y: &[T], groups: &[usize], n_groups: usize, ) -> Result<Self, CoreErrors>

Reduces a design and response to the sufficient statistics.

One O(n p^2) pass. Groups with no observations are dropped rather than counted, so a caller may pass a level set wider than the data.

Params

• x - Design matrix, n x p. Include the intercept column yourself.
• y - Response, length n
• groups - Level code per observation, each in 0..n_groups
• n_groups - Number of levels

Returns

The statistics, or CoreErrors::ShapeMismatch when the arguments disagree in length and CoreErrors::InvalidArgument for an out-of-range level code or fewer than two non-empty groups.

pub fn n_groups(&self) -> usize

Number of groups.

Returns

The count of non-empty levels.

Trait Implementations

impl Clone for RandomInterceptStats

fn clone(&self) -> Self

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for RandomInterceptStats

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.