ppforest2::Forest Class Reference

Abstract base class for projection pursuit random forests. More...

#include <Forest.hpp>

Public Types
using	FeatureMatrix = types::FeatureMatrix
using	FeatureVector = types::FeatureVector
using	Outcome = types::Outcome
using	OutcomeVector = types::OutcomeVector
using	Ptr = std::unique_ptr<Forest>
using	RNG = stats::RNG
Public Types inherited from ppforest2::Model
using	Ptr = std::shared_ptr<Model>

Public Member Functions
void	accept (Model::Visitor &visitor) const override=0
	Accept a model visitor (mode-specific dispatch).
void	add_tree (BaggedTree::Ptr tree)
	Add a trained bagged tree to the forest.
bool	operator!= (Forest const &other) const
bool	operator== (Forest const &other) const
Outcome	predict (FeatureVector const &x) const override=0
	Per-row prediction (mode-specific: majority vote or mean).
virtual types::OutcomeVector	predict (types::FeatureMatrix const &x) const
	Predict a matrix of observations.
Public Member Functions inherited from ppforest2::Model
virtual	~Model ()=default

Static Public Member Functions
static Ptr	train (TrainingSpec const &spec, FeatureMatrix const &x, OutcomeVector const &y)
	Train a random forest.
Static Public Member Functions inherited from ppforest2::Model
static void	check_train_inputs (types::FeatureMatrix const &x, types::OutcomeVector const &y)
	Validate common training inputs (y non-empty, matching x rows).
static Ptr	train (TrainingSpec const &spec, types::FeatureMatrix &x, types::OutcomeVector &y)
	Train a model from a training specification.

Public Attributes
std::vector< BaggedTree::Ptr >	trees
	Bootstrap-aggregated trees. Each BaggedTree pairs the polymorphic inner Tree with its sample indices.
Public Attributes inherited from ppforest2::Model
bool	degenerate = false
	Whether the model contains degenerate nodes/splits.
TrainingSpec::Ptr	training_spec
	Training specification used to build this model.

Protected Member Functions
	Forest ()=default
	Forest (TrainingSpec::Ptr spec)
void	build_trees (types::FeatureMatrix const &x, types::OutcomeVector const &y)
	Build training_spec->size bagged trees in parallel and attach them to this forest.
virtual BaggedTree::Ptr	train_tree (FeatureMatrix const &x, OutcomeVector const &y, RNG &rng) const =0
	Train one bagged tree on a bootstrap resample of x / y.

Detailed Description

Abstract base class for projection pursuit random forests.

Holds a vector of BaggedTree wrappers (each pairs a Tree with the bootstrap sample indices it was trained on) and the shared training spec. Aggregation logic (majority vote vs mean), proportion predictions, and OOB handling are defined in the concrete subclasses ClassificationForest and RegressionForest.

Construct via Forest::train, which dispatches to the correct concrete type based on training_spec.mode. Diagnostics and variable importance are free functions in models/Evaluation.hpp — they take a Forest const& (or Ptr) and dispatch on mode internally.

auto forest = Forest::train(spec, x, y);
auto preds  = forest->predict(x_test);

Member Typedef Documentation

FeatureMatrix

using ppforest2::Forest::FeatureMatrix = types::FeatureMatrix

FeatureVector

using ppforest2::Forest::FeatureVector = types::FeatureVector

Outcome

using ppforest2::Forest::Outcome = types::Outcome

OutcomeVector

using ppforest2::Forest::OutcomeVector = types::OutcomeVector

Ptr

using ppforest2::Forest::Ptr = std::unique_ptr<Forest>

RNG

using ppforest2::Forest::RNG = stats::RNG

Constructor & Destructor Documentation

Forest() [1/2]

ppforest2::Forest::Forest ( )

protecteddefault

Forest() [2/2]

ppforest2::Forest::Forest ( TrainingSpec::Ptr spec )

inlineexplicitprotected

Member Function Documentation

accept()

void ppforest2::Forest::accept ( Model::Visitor & visitor ) const

overridepure virtual

Accept a model visitor (mode-specific dispatch).

Implements ppforest2::Model.

Implemented in ppforest2::ClassificationForest, and ppforest2::RegressionForest.

add_tree()

void ppforest2::Forest::add_tree

(

BaggedTree::Ptr

tree

)

Add a trained bagged tree to the forest.

Asserts at runtime that the incoming tree's mode matches this forest's mode (e.g., a ClassificationForest can only accept trees trained under classification). This is the type-safety compromise of keeping Forest::trees mode-agnostic at the container level: the check runs at assembly time rather than at every prediction site, but is cheaper than threading templates through every call site that handles a Forest::Ptr.

build_trees()

void ppforest2::Forest::build_trees ( types::FeatureMatrix const & x, types::OutcomeVector const & y )

protected

Build training_spec->size bagged trees in parallel and attach them to this forest.

Shared training-loop scaffolding for ClassificationForest::train and RegressionForest::train. The two only differ in how a single bagged tree is trained — the outer scaffolding (size guard, OpenMP setup, per-tree retry loop with stream-id RNG, error capture, the post-loop assembly with degenerate-flag propagation) is identical. Mirrors the Tree::build_root pattern: shared algorithm lives on the base; the concrete subclass provides the mode-specific work via the train_tree virtual hook.

Each tree is trained using a RNG stream formula i + attempt * size, which is load-bearing for reproducibility.

Exceptions

Whatever

train_tree threw. If multiple iterations threw, the first (lowest-index) exception is rethrown.

operator!=()

bool ppforest2::Forest::operator!=

(

Forest const &

other

)

const

operator==()

bool ppforest2::Forest::operator==

(

Forest const &

other

)

const

predict() [1/2]

Outcome ppforest2::Forest::predict ( FeatureVector const & x ) const

overridepure virtual

Per-row prediction (mode-specific: majority vote or mean).

Implements ppforest2::Model.

Implemented in ppforest2::ClassificationForest, and ppforest2::RegressionForest.

predict() [2/2]

virtual types::OutcomeVector ppforest2::Model::predict ( types::FeatureMatrix const & x ) const

inlinevirtual

Predict a matrix of observations.

Default implementation iterates rows and dispatches to the single-row predict. Subclasses may override to vectorize.

Parameters

x	Feature matrix (n × p).

Returns

Predicted group labels (n).

Reimplemented from ppforest2::Model.

Reimplemented in ppforest2::RegressionForest.

train()

static Ptr ppforest2::Forest::train ( TrainingSpec const & spec, FeatureMatrix const & x, OutcomeVector const & y )

static

Train a random forest.

Dispatches to ClassificationForest::train or RegressionForest::train based on training_spec.mode. Note that the top-level x / y are not mutated here — each bootstrap tree resamples into its own local storage. (Contrast with single-tree Tree::train, where regression mode does permute x / y in place.)

train_tree()

virtual BaggedTree::Ptr ppforest2::Forest::train_tree ( FeatureMatrix const & x, OutcomeVector const & y, RNG & rng ) const

protectedpure virtual

Train one bagged tree on a bootstrap resample of x / y.

Mode-specific hook invoked by build_trees once per slot in the forest (with retries on degenerate trees). Subclasses that need per-training shared state (e.g. ClassificationForest caches the parent's label partition for stratified sampling) hold it as a private transient pointer set up before build_trees runs.

Called from inside build_trees's OpenMP parallel for, so implementations must be thread-safe and read-only over *this.

Implemented in ppforest2::ClassificationForest, and ppforest2::RegressionForest.

Member Data Documentation

trees

std::vector<BaggedTree::Ptr> ppforest2::Forest::trees

Bootstrap-aggregated trees. Each BaggedTree pairs the polymorphic inner Tree with its sample indices.

---

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

• core/src/models/Forest.hpp