Recovering Bandits

Authors: Ciara Pike-Burke, Steffen Grunewalder

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

Reproducibility Variable Result LLM Response
Research Type Experimental We complement these discussions with regret bounds and empirical studies. We show empirical results in Section 7
Researcher Affiliation Academia Ciara Pike-Burke Universitat Pompeu Fabra Barcelona, Spain c.pikeburke@gmail.com; Steffen Grünewälder Lancaster University Lancaster, UK s.grunewalder@lancaster.ac.uk
Pseudocode Yes Algorithm 1 d-step lookahead UCB and Thompson Sampling
Open Source Code No The paper references 'GPy: A gaussian process framework in python. http://github.com/Sheffield ML/GPy, 2012 .' which is a third-party library used for fitting GPs, not the source code for the authors' proposed algorithms. There is no other statement regarding the release of their code.
Open Datasets No The paper describes synthetic data generation: 'We tested our algorithms in experiments with zmax = 30, noise standard deviation σ = 0.1, and horizon T = 1000.' and 'sampled the recovery functions from a squared exponential kernel' or 'The recovery functions were logistic... and modified gamma'. No public dataset or access information is provided.
Dataset Splits No The paper describes the experimental setup including parameters like 'zmax = 30, noise standard deviation σ = 0.1, and horizon T = 1000.' and the number of replications, but it does not specify explicit training, validation, or test dataset splits.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running its experiments.
Software Dependencies No The paper mentions 'We used GPy [11] to fit the GPs.' and provides a reference to 'GPy: A gaussian process framework in python. http://github.com/Sheffield ML/GPy, 2012 .', but it does not specify a version number for GPy or any other software dependencies like Python, PyTorch, or TensorFlow.
Experiment Setup Yes We tested our algorithms in experiments with zmax = 30, noise standard deviation σ = 0.1, and horizon T = 1000. We averaged all results over 100 replications and used a squared exponential kernel with l = 4. We used squared exponential kernels in 1RGP-UCB and 1RGP-TS with lengthscale l = 5. The recovery functions were logistic, f(z) = θ0(1 + exp{ θ1(z θ2)}) 1 and modified gamma, f(z) = θ0C exp{ θ1z}zθ2