Struct LabelModel 

pub struct LabelModel { /* private fields */ }

Label Model for combining weak supervision sources.

Learns the accuracy and coverage of labeling functions to produce optimal probabilistic labels.

Example

use aprender::weak_supervision::{LabelModel, LFOutput};

// Matrix of LF outputs: [n_samples x n_labeling_functions]
let lf_matrix = vec![
    vec![LFOutput::Label(1), LFOutput::Abstain, LFOutput::Label(1)],
    vec![LFOutput::Label(0), LFOutput::Label(0), LFOutput::Abstain],
    // ...
];

let mut model = LabelModel::new(2, 3); // 2 classes, 3 LFs
model.fit(&lf_matrix, 100, 0.01);

let probs = model.predict_proba(&lf_matrix);

Implementations

impl LabelModel

pub fn new(n_classes: usize, n_lfs: usize) -> Self

Create a new label model.

Arguments

• n_classes - Number of output classes
• n_lfs - Number of labeling functions

pub fn fit(&mut self, lf_matrix: &[Vec<LFOutput>], n_epochs: usize, lr: f32)

Fit the label model using EM algorithm.

Arguments

• lf_matrix - Matrix of LF outputs [n_samples x n_lfs]
• n_epochs - Number of EM iterations
• lr - Learning rate for parameter updates

pub fn predict_proba(&self, lf_matrix: &[Vec<LFOutput>]) -> Vec<Vector<f32>>

Get probabilistic labels for samples.

pub fn predict(&self, lf_matrix: &[Vec<LFOutput>]) -> Vec<usize>

Get hard label predictions (argmax of probabilities).

pub fn get_accuracies(&self) -> &[Vec<f32>]

Get estimated LF accuracies.

pub fn get_class_priors(&self) -> &[f32]

Get estimated class priors.

pub fn get_lf_coverage(&self, lf_matrix: &[Vec<LFOutput>]) -> Vec<f32>

Get coverage of each LF (fraction of samples with non-abstain votes).

Trait Implementations

impl Clone for LabelModel

fn clone(&self) -> LabelModel

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for LabelModel

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

Formats the value using the given formatter.