Struct QcReport 

pub struct QcReport {
    pub train_keep: Vec<bool>,
    pub near_empty: Vec<bool>,
    pub feature_keep: Vec<bool>,
    pub n_genes: Vec<f32>,
    pub total_counts: Vec<f32>,
    pub mito_frac: Option<Vec<f32>>,
    pub ribo_frac: Option<Vec<f32>>,
    pub n_cells_dropped: usize,
    pub n_features_dropped: usize,
}

Per-cell / per-feature QC outcome.

Fields

train_keep: Vec<bool>

Cells to retain in training (len = num_columns). false only for non-near-empty MAD outliers — drives SparseIoVec::mask_columns.

near_empty: Vec<bool>

Near-empty cells (len = num_columns): kept in training, masked at output via QcReport::output_keep_idx.

feature_keep: Vec<bool>

Features to retain (len = num_rows). All true unless feature_min_cells > 0.

n_genes: Vec<f32>

total_counts: Vec<f32>

mito_frac: Option<Vec<f32>>

ribo_frac: Option<Vec<f32>>

n_cells_dropped: usize

n_features_dropped: usize

Implementations

impl QcReport

pub fn output_keep_idx(&self) -> Vec<usize>

Indices, in post-mask_columns column order, of the cells to emit at output: surviving (training-kept) cells that are not near-empty. Apply with Mat::select_rows on per-cell output matrices and filter the barcode slice by the same indices.

pub fn emit_idx_unmasked(&self) -> Vec<usize>

Emit-keep indices in the original (unmasked) column order: cells that are neither MAD-dropped nor near-empty. For paths that keep every cell in training and instead select_rows the per-cell outputs directly (e.g. bge’s UnifiedData, which has no mask_columns).