ppforest2::stats::GroupPartition Class Reference

Contiguous-block representation of grouped observations. More...

#include <GroupPartition.hpp>

Public Types
using	SplitSizes = std::map<types::GroupId, int>

Public Member Functions
	GroupPartition (int start, int end)
	Construct a single-group partition covering rows [start, end].
	GroupPartition (types::GroupIdVector const &y)
	Construct from a sorted response vector.
	GroupPartition (types::OutcomeVector const &y)
	Construct from a float-typed response vector.
types::FeatureMatrix	bgss (types::FeatureMatrix const &x) const
	Between-group sum of squares matrix (p × p).
GroupPartition	bisect (int mid) const
	Bisect a single-group partition at row index mid into two groups.
GroupPartition	collapse () const
	Collapse all groups into a single supergroup.
template<typename Derived>
auto	data (Eigen::MatrixBase< Derived > const &x) const
	Extract all rows across all groups.
Group	first_group () const
	Smallest group label in the partition.
template<typename Derived>
auto	group (Eigen::MatrixBase< Derived > const &x, Group const &group) const
	Extract rows belonging to a group (or supergroup).
int	group_end (Group const &group) const
	Last row index (inclusive) of the block for group.
int	group_size (Group const &group) const
	Number of observations in group.
int	group_start (Group const &group) const
	First row index of the block for group.
types::FeatureVector	mean (types::FeatureMatrix const &x) const
	Overall mean of all grouped rows (p).
GroupPartition	remap (GroupMap const &mapping) const
	Merge groups according to a mapping.
std::pair< GroupPartition, GroupPartition >	split (SplitSizes const &left_sizes) const
	Split each group's block into left and right children.
GroupPartition	subset (GroupSet const &groups) const
	Create a partition containing only the given groups.
int	total_size () const
	Total number of observations across all groups in the partition.
types::FeatureMatrix	wgss (types::FeatureMatrix const &x) const
	Within-group sum of squares matrix (p × p).

Static Public Member Functions
static bool	is_contiguous (GroupVector const &y)
	Check whether all equal values in y form a single contiguous block.

Public Attributes
GroupSet const	groups
	Set of all group labels in this partition.
GroupInvMap const	subgroups
	Maps each group to its set of subgroups.
GroupMap const	supergroups
	Maps each group to its supergroup (identity if no merge).

Detailed Description

Contiguous-block representation of grouped observations.

Assumes the response vector is sorted so that observations of the same group are contiguous. Stores the start/end row indices of each group block and provides efficient extraction, subsetting, and computation of between- and within-group statistics.

Groups can be hierarchically merged via remap(), which assigns supergroup labels while tracking the original subgroups.

// y must be sorted so equal values are contiguous.
GroupPartition y_part(y);
 
// Extract rows belonging to group 0:
auto x_group0 = y_part.group(x, 0);
 
// Between- and within-group statistics:
auto B = y_part.bgss(x);   // between-group sum of squares (p × p)
auto W = y_part.wgss(x);   // within-group sum of squares  (p × p)
 
// Restrict to a subset of groups:
GroupPartition sub = y_part.subset({0, 2});

Member Typedef Documentation

SplitSizes

using ppforest2::stats::GroupPartition::SplitSizes = std::map<types::GroupId, int>

Constructor & Destructor Documentation

GroupPartition() [1/3]

ppforest2::stats::GroupPartition::GroupPartition ( types::GroupIdVector const & y )

explicit

Construct from a sorted response vector.

Parameters

y	Outcome vector (n) with contiguous group blocks.

GroupPartition() [2/3]

ppforest2::stats::GroupPartition::GroupPartition ( types::OutcomeVector const & y )

explicit

Construct from a float-typed response vector.

Classification y is carried as OutcomeVector (float) throughout the training pipeline; this overload casts to integer labels internally before building the block map. Values must encode integer labels.

GroupPartition() [3/3]

ppforest2::stats::GroupPartition::GroupPartition	(	int	start,
		int	end )

Construct a single-group partition covering rows [start, end].

Group label is 0. Use bisect(mid) to split into a 2-group partition.

Member Function Documentation

bgss()

types::FeatureMatrix ppforest2::stats::GroupPartition::bgss

(

types::FeatureMatrix const &

x

)

const

Between-group sum of squares matrix (p × p).

bisect()

GroupPartition ppforest2::stats::GroupPartition::bisect

(

int

mid

)

const

Bisect a single-group partition at row index mid into two groups.

Group 0 covers [start, mid - 1], group 1 covers [mid, end]. The receiver must currently be a single-group partition (typically built via the (int, int) ctor). Distinct from split(SplitSizes) below, which subsets a multi-group partition along its existing structure.

Exceptions

via	invariant if the partition has more than one group, or if mid is outside (start, end].

collapse()

GroupPartition ppforest2::stats::GroupPartition::collapse

(

)

const

Collapse all groups into a single supergroup.

Returns

New GroupPartition with one supergroup containing all groups.

data()

template<typename Derived>

auto ppforest2::stats::GroupPartition::data ( Eigen::MatrixBase< Derived > const & x ) const

inline

Extract all rows across all groups.

Parameters

x	Feature matrix (n × p).

Returns

Sub-matrix with all grouped rows.

first_group()

Group ppforest2::stats::GroupPartition::first_group ( ) const

inline

Smallest group label in the partition.

groups is a std::set<GroupId> so iteration is in ascending key order; this returns the first such label. Caller must ensure the partition is non-empty.

group()

template<typename Derived>

auto ppforest2::stats::GroupPartition::group ( Eigen::MatrixBase< Derived > const & x, Group const & group ) const

inline

Extract rows belonging to a group (or supergroup).

Returns an Eigen block expression (zero-copy view) into x. The result must be consumed immediately or assigned to a concrete matrix — do not store it in auto across statements.

Parameters

x	Feature matrix (n × p).
group	Group label.

Returns

Block expression over the rows of group.

group_end()

int ppforest2::stats::GroupPartition::group_end

(

Group const &

group

)

const

Last row index (inclusive) of the block for group.

group_size()

int ppforest2::stats::GroupPartition::group_size

(

Group const &

group

)

const

Number of observations in group.

group_start()

int ppforest2::stats::GroupPartition::group_start

(

Group const &

group

)

const

First row index of the block for group.

is_contiguous()

static bool ppforest2::stats::GroupPartition::is_contiguous ( GroupVector const & y )

static

Check whether all equal values in y form a single contiguous block.

mean()

types::FeatureVector ppforest2::stats::GroupPartition::mean

(

types::FeatureMatrix const &

x

)

const

Overall mean of all grouped rows (p).

remap()

GroupPartition ppforest2::stats::GroupPartition::remap

(

GroupMap const &

mapping

)

const

Merge groups according to a mapping.

Parameters

mapping

Maps original group labels to supergroup labels.

Returns

New GroupPartition with merged groups.

split()

std::pair< GroupPartition, GroupPartition > ppforest2::stats::GroupPartition::split

(

SplitSizes const &

left_sizes

)

const

Split each group's block into left and right children.

For each leaf group, left_sizes specifies how many rows go to the left child (the first rows of the block). The remaining rows go to the right child. Groups absent from left_sizes go entirely to the right child (left_count = 0).

The caller is responsible for having already reordered rows within each block so that left-bound observations come first.

Parameters

left_sizes

Maps each leaf group to its left child row count.

Returns

Pair of {left, right} GroupPartitions.

subset()

GroupPartition ppforest2::stats::GroupPartition::subset

(

GroupSet const &

groups

)

const

Create a partition containing only the given groups.

Parameters

groups

Set of group labels to keep.

Returns

New GroupPartition restricted to groups.

total_size()

int ppforest2::stats::GroupPartition::total_size

(

)

const

Total number of observations across all groups in the partition.

Used by the tree builder to detect "no-progress" grouping splits: if a child partition covers the same row count as its parent, the split failed to partition the data and the builder converts the node to a leaf to avoid unbounded recursion.

wgss()

types::FeatureMatrix ppforest2::stats::GroupPartition::wgss

(

types::FeatureMatrix const &

x

)

const

Within-group sum of squares matrix (p × p).

Member Data Documentation

groups

GroupSet const ppforest2::stats::GroupPartition::groups

Set of all group labels in this partition.

subgroups

GroupInvMap const ppforest2::stats::GroupPartition::subgroups

Maps each group to its set of subgroups.

supergroups

GroupMap const ppforest2::stats::GroupPartition::supergroups

Maps each group to its supergroup (identity if no merge).

---

The documentation for this class was generated from the following file:

• core/src/stats/GroupPartition.hpp