Bandit Multiclass Linear Classification: Efficient Algorithms for the Separable Case

Authors: Alina Beygelzimer, David Pal, Balazs Szorenyi, Devanathan Thiruvenkatachari, Chen-Yu Wei, Chicheng Zhang

ICML 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we provide an empirical evaluation on our algorithms, verifying their effectiveness on linearly separable datasets. We generated strongly and weakly linearly separable datasets with K = 3 classes in R3 i.i.d. from two data distributions. Figures 3a and 3b show visualizations of the two datasets, along with detailed descriptions of the distributions.
Researcher Affiliation Collaboration 1Yahoo Research, New York, NY, USA 2New York University, New York, NY, USA 3University of Southern California, Los Angeles, CA, USA 4Microsoft Research, New York, NY, USA.
Pseudocode Yes Algorithm 1 BANDIT ALGORITHM FOR STRONGLY LINEARLY SEPARABLE EXAMPLES...Algorithm 2 KERNELIZED BANDIT ALGORITHM
Open Source Code No The paper mentions a MATLAB toolbox by Orabona (2009) as being available, but this is an external tool and not the open-source code for the specific algorithms and methods presented in this paper.
Open Datasets No We generated strongly and weakly linearly separable datasets with K = 3 classes in R3 i.i.d. from two data distributions. Figures 3a and 3b show visualizations of the two datasets, along with detailed descriptions of the distributions.
Dataset Splits No The paper states details about the generated datasets, such as the number of classes, dimensionality, and class distribution (e.g., "80% of the examples belong to class 1, 10% belong to class 2 and 10% belong to class 3."). However, it does not specify any train, validation, or test splits, nor does it mention cross-validation or other explicit data partitioning methodologies required for reproducibility.
Hardware Specification No The paper does not provide any specific details regarding the hardware used for running the experiments, such as CPU or GPU models, memory, or cloud instance types.
Software Dependencies No The paper mentions implementing algorithms and using the BANDITRON algorithm by Orabona (2009), which is a MATLAB toolbox. However, it does not list specific software dependencies with version numbers (e.g., Python 3.x, PyTorch 1.x, specific library versions) that would be needed to reproduce the experiments.
Experiment Setup Yes BANDITRON has an exploration rate parameter, for which we tried values 0.02, 0.01, 0.005, 0.002, 0.001, 0.0005.