Recovering True Classifier Performance in Positive-Unlabeled Learning

Authors: Shantanu Jain, Martha White, Predrag Radivojac

AAAI 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Using state-of-the-art algorithms to estimate the positive class prior and the proportion of noise, we experimentally evaluate two correction approaches and demonstrate their efficacy on real-life data.
Researcher Affiliation Academia Shantanu Jain, Martha White, Predrag Radivojac Department of Computer Science Indiana University, Bloomington, Indiana, USA {shajain, martha, predrag}@indiana.edu
Pseudocode No The paper states 'The full algorithm for the indirect recovery is given in the ar Xiv supplement of this paper.' but no pseudocode or algorithm blocks are present in the provided text.
Open Source Code No The paper does not provide any explicit statement about releasing source code or a link to a code repository for the methodology described.
Open Datasets Yes Our estimators were evaluated using twelve real-life data sets from the UCI Machine Learning Repository (Lichman 2013).
Dataset Splits Yes A validation set containing 25% of the training data was used to terminate training.The number of actual positive examples in each labeled set was a function of parameter β {1, 0.95, 0.75}.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running experiments.
Software Dependencies No The paper mentions algorithms used (feedforward neural networks, resilient propagation) but does not provide specific software names with version numbers for dependencies (e.g., Python, PyTorch, TensorFlow versions).
Experiment Setup Yes Classifiers were constructed as ensembles of 100 feedforward neural networks... Each network had five hidden neurons and was trained using resilient propagation... A validation set containing 25% of the training data was used to terminate training.