Abstract strategy for managing group partitions. More...

#include <Grouping.hpp>

Public Member Functions
stats::GroupPartition	init (types::GroupIdVector const &y) const
	Convenience overload for callers with a `GroupIdVector`.

stats::GroupPartition	init (types::OutcomeVector const &y) const
	Create the initial GroupPartition from the training response.

void	split (NodeContext &ctx, stats::RNG &rng) const
	Split observations into two child partitions; writes ctx.lower_y_part / upper_y_part.

Public Member Functions inherited from Strategy< Grouping >
virtual	~Strategy ()=default

virtual std::string	display_name () const=0
	Human-readable name for display in summaries.

virtual std::set< ppforest2::types::Mode >	supported_modes () const=0
	Training modes this strategy supports.

virtual nlohmann::json	to_json () const=0
	Serialize this strategy's configuration to JSON.

Protected Member Functions
virtual void	compute (NodeContext &ctx, stats::RNG &rng) const =0
	Subclass implementation of per-node splitting.

virtual stats::GroupPartition	compute_init (types::OutcomeVector const &y) const =0
	Subclass implementation of initial partition construction.

Additional Inherited Members
Public Types inherited from Strategy< Grouping >
using	Factory
	Factory function type for deserializing a strategy from JSON.

using	Ptr

Static Public Member Functions inherited from Strategy< Grouping >
static Ptr	from_json (nlohmann::json const &j)
	Construct a strategy from its JSON representation.

static bool	register_strategy (std::string const &name, Factory factory)
	Register a concrete strategy for JSON deserialization.

Detailed Description

Abstract strategy for managing group partitions.

init() builds the root partition once before training; split() writes ctx.lower_y_part and ctx.upper_y_part per node.

Member Function Documentation

◆ compute()

virtual void ppforest2::grouping::Grouping::compute	(	NodeContext &	ctx,
		stats::RNG &	rng ) const

protectedpure virtual

Subclass implementation of per-node splitting.

Implemented in ppforest2::grouping::ByCutpoint, and ppforest2::grouping::ByLabel.

◆ compute_init()

virtual stats::GroupPartition ppforest2::grouping::Grouping::compute_init ( types::OutcomeVector const & y ) const

protectedpure virtual

Subclass implementation of initial partition construction.

Implemented in ppforest2::grouping::ByCutpoint, and ppforest2::grouping::ByLabel.

◆ init() [1/2]

stats::GroupPartition ppforest2::grouping::Grouping::init ( types::GroupIdVector const & y ) const

inline

Convenience overload for callers with a GroupIdVector.

◆ init() [2/2]

stats::GroupPartition ppforest2::grouping::Grouping::init ( types::OutcomeVector const & y ) const

Create the initial GroupPartition from the training response.

Caller must pre-sort rows so equal values (classification) or ascending response (regression) form contiguous blocks.

Public NVI entry point: delegates to the subclass-provided compute_init, then asserts the partition's post-conditions — blocks rooted at row 0 and covering all of y. Downstream callers (e.g. stratified resampling in ClassificationForest::train_tree) rely on those properties to safely reuse the parent partition over a resampled response.

◆ split()

void ppforest2::grouping::Grouping::split	(	NodeContext &	ctx,
		stats::RNG &	rng ) const

Split observations into two child partitions; writes ctx.lower_y_part / upper_y_part.

Public NVI entry point: skips if ctx.aborted is set, otherwise delegates to the subclass-provided compute.

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

core/src/models/strategies/grouping/Grouping.hpp

Public Member Functions

Protected Member Functions