An Equivalence Analysis of Binary Quantification Methods

Authors: Alberto Castaño, Jaime Alonso, Pablo González, Juan José del Coz

AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental After an empirical evaluation of all these methods using synthetic and benchmark datasets, the paper concludes recommending three of them due to their precision, efficiency, and diversity.
Researcher Affiliation Academia Alberto Casta no, Jaime Alonso, Pablo Gonz alez, Juan Jos e del Coz Artificial Intelligence Center University of Oviedo. Campus de Viesques, 33204, Gij on, Spain uo226218@uniovi.es, jalonso@uniovi.es, gonzalezgpablo@uniovi.es, juanjo@uniovi.es
Pseudocode Yes Algorithm 1: Mixture function used by QUANTy
Open Source Code Yes This study1 compares six of the quantification algorithms previously discussed: AC, HDy, PAC, QUANTy, SORD, and MM (renamed as CDFy) and has two goals. (...) 1github.com/bertocast/binary-quantification-equivalence
Open Datasets Yes After an empirical evaluation of all these methods using synthetic and benchmark datasets, the paper concludes recommending three of them due to their precision, efficiency, and diversity. (...) The second group of experiments was carried out using benchmark datasets.
Dataset Splits Yes Following (Forman 2008), 50-fold CV was used to estimate the training distributions. (...) The RF hyperparameters (depth, number of trees and minimum number of examples for the leaf nodes) were automatically adjusted using a grid search and 3fold cross-validation optimizing the geometric mean to obtain adequate classifiers even when classes were unbalanced. All the quantifiers were trained over the same partitions, 70% for training and 30% for testing, with 40 repetitions.
Hardware Specification Yes The authors gratefully acknowledge the support of NVIDIA Corporation with the donation of the Titan Xp GPU used for this research.
Software Dependencies No The paper mentions 'Logistic Regression' and 'Random Forest' as base learners, but does not provide specific version numbers for these libraries or any other key software components, such as Python or PyTorch/TensorFlow versions.
Experiment Setup Yes Logistic Regression with C =1 was employed to train the binary classifiers. (...) The RF hyperparameters (depth, number of trees and minimum number of examples for the leaf nodes) were automatically adjusted using a grid search and 3fold cross-validation optimizing the geometric mean to obtain adequate classifiers even when classes were unbalanced.