Struct SparseIoStack 

pub struct SparseIoStack {
    pub stack: Vec<SparseIoVec>,
    /* private fields */
}

sparse_io_stack is a stack of sparse_io_vector

Fields

stack: Vec<SparseIoVec>

Implementations

impl SparseIoStack

pub fn new() -> Self

pub fn push(&mut self, data: SparseIoVec) -> Result<()>

pub fn num_types(&self) -> usize

number of data types

pub fn num_columns(&self) -> Result<usize>

number of shared columns

pub fn column_names(&self) -> Result<Vec<Box<str>>>

Get the shared column names.

pub fn mask_columns_all(&mut self, keep: &[bool]) -> Result<()>

Exclude columns (cells) across every layer in the stack in lockstep. keep[global_col] is true to keep, false to drop. Mirror of SparseIoVec::mask_columns; the members share one synchronized cell axis, so the same mask applies to all. MUST be called before batch/group registration.

pub fn register_batch_membership<T>(&mut self, batch_membership: &[T])

where T: Sync + Send + Hash + Eq + Clone + ToString,

Register batch membership for all layers in the stack.

Panics

If batch_membership is not one entry per column; try_register_batch_membership returns that as an error instead.

pub fn try_register_batch_membership<T>( &mut self, batch_membership: &[T], ) -> Result<()>

where T: Sync + Send + Hash + Eq + Clone + ToString,

register_batch_membership that returns a membership of the wrong length as an error. The layers share their columns, so a refused membership is registered on none.

pub fn row_names(&self) -> Result<Vec<Box<str>>>

Get the row names combined across all the types. We will append additional data type index: format!("{}/{}", x, d)

Trait Implementations

impl Default for SparseIoStack

fn default() -> Self

an empty sparse io vector for horizontal data integration

impl MultilevelCollapsingOps for SparseIoStack

type LevelOutput = Vec<CollapsedOut>

fn collapse_columns_multilevel<T>( &mut self, proj_kn: &DMatrix<f32>, batch_membership: &[T], params: &MultilevelParams, ) -> Result<Vec<CollapsedOut>>

where T: Sync + Send + Hash + Eq + Clone + ToString,

fn collapse_columns_multilevel_vec<T>( &mut self, proj_kn: &DMatrix<f32>, batch_membership: &[T], params: &MultilevelParams, ) -> Result<Vec<Vec<CollapsedOut>>>

where T: Sync + Send + Hash + Eq + Clone + ToString,

impl RandProjOps for SparseIoStack

fn project_columns_with_batch_correction_seeded<T>( &self, target_dim: usize, block_size: Option<usize>, batch_membership: Option<&[T]>, seed: u64, ) -> Result<RandColProjOut>

where T: Sync + Send + Hash + Eq + Clone + ToString,

Seeded variant of RandProjOps::project_columns_with_batch_correction: seed pins the random basis so identical-config runs are byte-identical.

fn project_columns_weighted_seeded<T>( &self, target_dim: usize, block_size: Option<usize>, batch_membership: Option<&[T]>, row_weights: &[f32], seed: u64, ) -> Result<RandColProjOut>

where T: Sync + Send + Hash + Eq + Clone + ToString,

Seeded variant of RandProjOps::project_columns_weighted: seed pins the random basis so identical-config runs are byte-identical.

fn partition_columns_to_groups( &mut self, proj_kn: &DMatrix<f32>, num_sorting_features: Option<usize>, ncols_per_group: Option<usize>, ) -> Result<usize>

Assign each column/cell to random group by binary encoding

fn assign_group_labels(&mut self, labels: &[usize])

Assign each column to the group of its label.

fn project_columns( &self, target_dim: usize, block_size: Option<usize>, ) -> Result<RandColProjOut>

Create K x ncol projection by concatenating data across columns and returns the random basis matrix D x target_dim and the projection result target_dim x N

fn project_columns_with_batch_correction<T>( &self, target_dim: usize, block_size: Option<usize>, batch_membership: Option<&[T]>, ) -> Result<RandColProjOut>

where T: Sync + Send + Hash + Eq + Clone + ToString,

Create K x ncol projection by concatenating data across columns and returns the random basis matrix D x target_dim and the projection result target_dim x N

fn project_columns_weighted<T>( &self, target_dim: usize, block_size: Option<usize>, batch_membership: Option<&[T]>, row_weights: &[f32], ) -> Result<RandColProjOut>

where T: Sync + Send + Hash + Eq + Clone + ToString,

Like project_columns_with_batch_correction but with per-row (per-feature) weights applied to the random basis. Features with weight 0 are excluded from the projection geometry.

fn partition_columns_to_mixed_groups<T>( &mut self, proj_kn: &DMatrix<f32>, num_features: Option<usize>, batch_membership: &[T], min_batches: usize, merge_levels: usize, ) -> Result<usize>

where T: Hash + Eq + Clone,

Like RandProjOps::partition_columns_to_groups, but bins that hold fewer than min_batches batches merge up the code tree, for at most merge_levels levels (see crate::alg::batch_mixing). min_batches is capped by the number of batches. Returns the number of groups.