Struct PbSampleLayout 

pub struct PbSampleLayout {
    pub centroids: DMatrix<f32>,
    pub cell_counts: Vec<f32>,
    pub pb_sample_to_batch: Vec<usize>,
    pub pb_sample_to_group: Vec<usize>,
    pub bg_to_pbsamp: FxHashMap<(usize, usize), usize>,
    pub cell_to_pbsamp: Vec<usize>,
    pub singleton_col: Vec<Option<usize>>,
    pub bulk_batches: Vec<usize>,
    pub pb_sample_to_stratum: Option<Vec<usize>>,
}

Fields

centroids: DMatrix<f32>

Centroid matrix: proj_dim x num_pb_samples

cell_counts: Vec<f32>

Number of cells in each pb-sample

pb_sample_to_batch: Vec<usize>

Batch index for each pb-sample

pb_sample_to_group: Vec<usize>

Sample/group index for each pb-sample

bg_to_pbsamp: FxHashMap<(usize, usize), usize>

Maps (batch, group) → pb-sample index

cell_to_pbsamp: Vec<usize>

Global cell index → owning pb-sample index.

singleton_col: Vec<Option<usize>>

For a singleton pb-sample (an anchored reference column or a bulk sample), the one column it stands for; None for ordinary (batch, group) pb-samples. The layout is the single owner of “which pb-samples are already summaries” — downstream steps (split_anchored_finest_groups) read it here instead of re-deriving membership from batch labels.

bulk_batches: Vec<usize>

Batches whose columns are bulk samples: singletons like anchors, but excluded from cross-batch matching in BOTH directions (see [bbknn_match_one_pbsamp]).

pb_sample_to_stratum: Option<Vec<usize>>

Per-pb-sample CNV stratum when collapse is stratified. Same-stratum BBKNN filter reads this; None disables the filter.

Implementations

impl PbSampleLayout

pub fn is_bulk(&self, pbsamp: usize) -> bool

Whether this pb-sample is a bulk sample (a singleton that takes no part in matching).

pub fn is_bulk_batch(&self, batch: usize) -> bool

Whether batch is a bulk batch — the one spelling of the predicate, so per-pb-sample and per-batch callers cannot drift apart.