Struct LinearClassifier 

pub struct LinearClassifier {
    pub hidden: usize,
    pub num_classes: usize,
    /* private fields */
}

Trained [num_classes, hidden] weight matrix + per-class bias.

Fields

hidden: usize

num_classes: usize

Implementations

impl LinearClassifier

pub fn new(hidden: usize, num_classes: usize) -> LinearClassifier

Initialize weights with a small Gaussian scale (Xavier-ish).

pub fn predict(&self, features: &[f32]) -> Result<usize, Error>

Predict the argmax class for one example.

pub fn accuracy(&self, examples: &[LabeledFeature<'_>]) -> Result<f32, Error>

Batched argmax accuracy on a frozen dataset.