5.2. Metrics and scoring: quantifing the performance of a quantifier

The measures implemented in this module are shown in the book Learning to Quantifier, and can be listed as follows:

measure	abbreviation	First proposed for quantification	Return type
Absolute error	ae	–	array-like
Mean Absolute Error	mae	Saerens et al. (2002)	float
Normalized Absolute Error	nae	Esuli and Sebastiani (2014)	float
Relative Absolute Error	rae	González-Castro et al. (2010)	float
Normalized Relative Absolute Error	nrae	Esuli and Sebastiani (2014)	float
Squared Error	se	Bella et al. (2011)	float
Mean Squared Error	mse	–	float
Kullback Leibler Divergence	kld	Forman (2005)	array-like
Normalized Kullback Leibler Divergence	nkld	Esuli and Sebastiani (2014)	float

Note

Return type

When inserting the measure in the scoring parameter of GridSearchQ or APP classes, the name of the measure should be passed with its acronym, for example “mae” for mean_absolute_error.