Struct HardNegativeMiner 

pub struct HardNegativeMiner {
    pub strategy: MiningStrategy,
    pub pos_neg_ratio: f64,
}

Hard negative mining for handling imbalanced datasets.

Focuses training on difficult negative examples to improve classifier discrimination.

References

• Shrivastava et al. (2016): “Training Region-based Object Detectors with Online Hard Example Mining”

Fields

strategy: MiningStrategy

Mining strategy to use

pos_neg_ratio: f64

Ratio of positives to negatives to maintain

Implementations

impl HardNegativeMiner

pub fn new(strategy: MiningStrategy, pos_neg_ratio: f64) -> Self

Create a new hard negative miner.

pub fn select_samples( &self, losses: &Array1<f64>, labels: &Array1<f64>, ) -> TrainResult<Vec<usize>>

Select hard negative samples based on loss values.

Arguments

• losses - Per-sample loss values
• labels - True labels (1.0 for positive, 0.0 for negative)

Returns

Indices of selected samples

Trait Implementations

impl Clone for HardNegativeMiner

fn clone(&self) -> HardNegativeMiner

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for HardNegativeMiner

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

Formats the value using the given formatter.