Package deepnetts.eval

Class ClassifierEvaluator

java.lang.Object

deepnetts.eval.ClassifierEvaluator

All Implemented Interfaces:

javax.visrec.ml.eval.Evaluator<NeuralNetwork,javax.visrec.ml.data.DataSet<? extends MLDataItem>>

public class ClassifierEvaluator
extends Object
implements javax.visrec.ml.eval.Evaluator<NeuralNetwork,javax.visrec.ml.data.DataSet<? extends MLDataItem>>

Evaluation method for binary and multi-class classifiers. Calculates classification performance metrics, how good is it at predicting class of something. http://www.ritchieng.com/machine-learning-evaluate-classification-model/ http://scikit-learn.org/stable/modules/model_evaluation.html http://notesbyanerd.com/2014/12/17/multi-class-performance-measures/ https://en.wikipedia.org/wiki/Confusion_matrix https://stats.stackexchange.com/questions/21551/how-to-compute-precision-recall-for-multiclass-multilabel-classification http://scikit-learn.org/stable/modules/model_evaluation.html#confusion-matrix Micro and macro averaging: https://datascience.stackexchange.com/questions/15989/micro-average-vs-macro-average-performance-in-a-multiclass-classification-settin I'm doing macro

Constructor Summary

Constructors

Constructor	Description
ClassifierEvaluator()

Method Summary

Modifier and Type	Method	Description
ClassificationMetrics	evaluate(NeuralNetwork neuralNet, javax.visrec.ml.data.DataSet<? extends MLDataItem> testSet)	Performs classifier evaluation and returns classification performance metrics.
ConfusionMatrix	getConfusionMatrix()
ClassificationMetrics	getMacroAverage()
Map<String,ClassificationMetrics>	getMetricsByClass()
float	getThreshold()
static javax.visrec.ml.eval.EvaluationMetrics	macroAverage(Collection<javax.visrec.ml.eval.EvaluationMetrics> metrics)	Calculates macro average for the given list of ClassificationMetrics.
void	setThreshold(float threshold)
String	toString()

Methods inherited from class java.lang.Object

equals, getClass, hashCode, notify, notifyAll, wait, wait, wait

Constructor Details

ClassifierEvaluator

public ClassifierEvaluator()

Method Details

evaluate

public ClassificationMetrics evaluate(NeuralNetwork neuralNet,
 javax.visrec.ml.data.DataSet<? extends MLDataItem> testSet)

Performs classifier evaluation and returns classification performance metrics.

Specified by:

evaluate in interface javax.visrec.ml.eval.Evaluator<NeuralNetwork,javax.visrec.ml.data.DataSet<? extends MLDataItem>>

Parameters:

neuralNet -

testSet -

Returns:

getThreshold

public float getThreshold()

setThreshold

public void setThreshold(float threshold)

getMacroAverage

public ClassificationMetrics getMacroAverage()

macroAverage

public static javax.visrec.ml.eval.EvaluationMetrics macroAverage(Collection<javax.visrec.ml.eval.EvaluationMetrics> metrics)

Calculates macro average for the given list of ClassificationMetrics. Used for multi-class classification. Its suitable for balanced classes, but if not micro averaging is more suitable. Micro averaging is summing confusion matrices for individual classes.

Parameters:

metrics -

Returns:

getMetricsByClass

public Map<String,ClassificationMetrics> getMetricsByClass()

getConfusionMatrix

public ConfusionMatrix getConfusionMatrix()

toString

public String toString()

Overrides:

toString in class Object