drwnClassificationResults Class Reference

Encapsulates summary of classifier output from which various curves can be generated (e.g., precision-recall curves). More...

Inheritance diagram for drwnClassificationResults:

Public Member Functions
	drwnClassificationResults ()
	default constructor
	drwnClassificationResults (const drwnClassificationResults &c)
	copy constructor
int	numPositives () const
	return the number os positive samples accumulated
int	numNegatives () const
	return the number of negative samples accumulated
int	numSamples () const
	return the total number (positive and negative) of samples accumulated
int	numThresholds () const
	return the number of unique classification scores
int	numMisses () const
	return the number of positive samples that have not been scored
double	getPosWeight () const
	return the relative weight of a positive sample to a negative sample
void	setPosWeight (double w)
	set the relative weight of a positive sample to a negative sample
void	normalize ()
	this will change the weight of the positive examples such that overall positive and negative examples will have the same weight
void	clear ()
	clear the accumulated scores
bool	write (const char *filename) const
	write the accumulated scores to file
bool	read (const char *filename)
	read accumulated scores from file
void	accumulate (const drwnClassificationResults &c)
	accumulate results from another drwnClassificationResults object
void	accumulate (const drwnClassifierDataset &dataset, drwnClassifier const *classifier, int positiveClassId=1)
	Accumulate results from a classifier run on a dataset. The positiveClass parameter indicates the positive class label for multi-class classifiers.
void	accumulatePositives (double score, int count=1)
	accumulate a single positive example
void	accumulatePositives (const vector< double > &scores)
	accumulate multiple positive examples
void	accumulateNegatives (double score, int count=1)
	accumulate a single negative example
void	accumulateNegatives (const vector< double > &scores)
	accumulate multiple negative examples
void	accumulateMisses (int count=1)
	accumulate unscored positive examples (misses)

Static Public Attributes
static bool	INCLUDE_MISSES = false
	true if some positive samples are never scored

Protected Attributes
map< double, pair< int, int > >	_scoredResults
	number of positives (first) and negatives (second) grouped by score
int	_numPositiveSamples
	must be greater than sum(_scoredResults.first)
int	_numNegativeSamples
	must be must be equal to sum(_scoredResults.second)
double	_posWeight
	weight of positive-to-negative count

Detailed Description

Encapsulates summary of classifier output from which various curves can be generated (e.g., precision-recall curves).

The results can be scored arbitrarily so long as higher score implies more likely. Positive and negative samples can be weighted differently. The following code snippet shows how to produce a precision-recall curve that can be subsequently plotted in Matlab.

vector<int> y = getGroundTruthLabels();
vector<double> y_hat = getPredictionProbabilities();

drwnPRCurve curve;
for (int n = 0; n < N; n++) {
    if (y[n] == 1) curve.accumulatePositives(y_hat[n]);
    else curve.accumulateNegatives(y_hat[n]);
}

curve.writeCurve("pr.txt");

See Also

drwnPRCurve, drwnML Tutorial

---

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

• drwnClassificationResults.h
• drwnClassificationResults.cpp