FST::Branch_And_Bound_Predictor_Averaging< RETURNTYPE, DIMTYPE > Class Template Reference

Averaging Prediction Mechanism in Fast Branch and Bound & Branch and Bound with Partial Prediction. More...

#include <branch_and_bound_predictor_averaging.hpp>

Inheritance diagram for FST::Branch_And_Bound_Predictor_Averaging< RETURNTYPE, DIMTYPE >:

Public Member Functions
virtual void	initialize (const DIMTYPE features) override
	(re)initializes the per-feature sum/count accumulators to zero for features features; call before use for each new search
virtual bool	learn (const DIMTYPE feature, const RETURNTYPE value) override
	adds one observed value (non-negative, asserted) to feature 's running sum and increments its sample count; always returns true
virtual bool	predict (const DIMTYPE feature, RETURNTYPE &value) const override
	sets value to the gamma-scaled running average learned for feature and returns true, but only once at least _delta samples have been learned for it; returns false otherwise
unsigned long	get_count (const DIMTYPE feature) const
	the number of learn() calls recorded so far for feature - a diagnostic accessor, telling how well founded the feature's prediction is (predict() succeeds once this reaches the delta set by set_delta())
void	set_gamma (const RETURNTYPE gamma=1.0)
	sets the scaling factor applied to the averaged value in predict() (asserted >=0.0); default 1.0 = plain average
void	set_delta (const unsigned int delta=1)
	sets the minimum sample count a feature must reach before predict() returns a value (asserted >=1); default 1 = predict as soon as one sample is learned
virtual std::ostream &	print (std::ostream &os) const override
	writes gamma/delta plus, for every feature with at least one learned sample, its current predicted value and sample count, and the total sample count across all features

Protected Attributes
std::vector< RETURNTYPE >	_value
	sums up 'learned' values
std::vector< unsigned long >	_count
	records the number of summed up values
RETURNTYPE	_gamma
	scaling factor applied to the averaged prediction; see set_gamma()
unsigned int	_delta
	minimum sample count before predict() returns a value; see set_delta()

Detailed Description

template<class RETURNTYPE, typename DIMTYPE>
class FST::Branch_And_Bound_Predictor_Averaging< RETURNTYPE, DIMTYPE >

Averaging Prediction Mechanism in Fast Branch and Bound & Branch and Bound with Partial Prediction.

Caller convention (see Search_Branch_And_Bound_Fast/_Partial_Prediction): the "value" learned for a feature is the nonnegative criterion-value drop caused by removing that feature from a node's parent (parent value - child value), and the "value" predicted for a feature is used the same way to estimate an unevaluated child's value as parent value - predicted drop. This class itself is agnostic to that interpretation; it just accumulates and averages whatever per-feature values it is given.

Member Function Documentation

initialize()

template<class RETURNTYPE, typename DIMTYPE>

void FST::Branch_And_Bound_Predictor_Averaging< RETURNTYPE, DIMTYPE >::initialize ( const DIMTYPE features )

overridevirtual

(re)initializes the per-feature sum/count accumulators to zero for features features; call before use for each new search

Implements FST::Branch_And_Bound_Predictor< RETURNTYPE, DIMTYPE >.

learn()

template<class RETURNTYPE, typename DIMTYPE>

bool FST::Branch_And_Bound_Predictor_Averaging< RETURNTYPE, DIMTYPE >::learn ( const DIMTYPE feature, const RETURNTYPE value )

overridevirtual

adds one observed value (non-negative, asserted) to feature 's running sum and increments its sample count; always returns true

Implements FST::Branch_And_Bound_Predictor< RETURNTYPE, DIMTYPE >.

predict()

template<class RETURNTYPE, typename DIMTYPE>

bool FST::Branch_And_Bound_Predictor_Averaging< RETURNTYPE, DIMTYPE >::predict ( const DIMTYPE feature, RETURNTYPE & value ) const

overridevirtual

sets value to the gamma-scaled running average learned for feature and returns true, but only once at least _delta samples have been learned for it; returns false otherwise

Implements FST::Branch_And_Bound_Predictor< RETURNTYPE, DIMTYPE >.

print()

template<class RETURNTYPE, typename DIMTYPE>

std::ostream & FST::Branch_And_Bound_Predictor_Averaging< RETURNTYPE, DIMTYPE >::print ( std::ostream & os ) const

overridevirtual

writes gamma/delta plus, for every feature with at least one learned sample, its current predicted value and sample count, and the total sample count across all features

Reimplemented from FST::Branch_And_Bound_Predictor< RETURNTYPE, DIMTYPE >.

---

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

• /Users/psms/dev/devtest1/_src_search/branch_and_bound_predictor_averaging.hpp