Struct ConditionalLinear 

pub struct ConditionalLinear {
    pub node_idx: usize,
    pub w: Vec<Vec<f64>>,
    pub b: Vec<f64>,
    pub sigma: Vec<f64>,
    pub n_classes: usize,
    pub parent_indices: Vec<usize>,
}

Conditional Linear Gaussian for continuous parents and discrete output.

P(X=k | pa(X)) = softmax(W[k] * pa(X) + b[k]) sigma[k] stores standard deviations (for density evaluation).

Fields

node_idx: usize

Index of this node.

w: Vec<Vec<f64>>

Weight matrix: W[k] has length = number of parents.

b: Vec<f64>

Bias vector: b[k] for each class k.

sigma: Vec<f64>

Standard deviations (used if output is also continuous).

n_classes: usize

Number of output classes.

parent_indices: Vec<usize>

Parent node indices.

Implementations

impl ConditionalLinear

pub fn new( node_idx: usize, w: Vec<Vec<f64>>, b: Vec<f64>, sigma: Vec<f64>, n_classes: usize, parent_indices: Vec<usize>, ) -> Result<Self, StatsError>

Create a new ConditionalLinear CPD.

pub fn softmax(&self, parent_values: &[f64]) -> Vec<f64>

Compute softmax probabilities.

Trait Implementations

impl CPD for ConditionalLinear

fn node(&self) -> usize

Name of the node this CPD belongs to.

fn prob(&self, value: usize, parent_values: &[usize]) -> f64

Evaluate the (unnormalised) probability (or density) of value given parent_values.

fn cardinality(&self) -> usize

Number of states (cardinality) for discrete nodes; 0 for continuous.

fn parent_indices(&self) -> &[usize]

Parent node indices.

fn is_continuous(&self) -> bool

Whether this CPD is continuous.

fn log_prob(&self, value: usize, parent_values: &[usize]) -> f64

Log-probability: default implementation wraps prob.

impl Clone for ConditionalLinear

fn clone(&self) -> ConditionalLinear

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for ConditionalLinear

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

Formats the value using the given formatter.