An optimal algorithm for the Thresholding Bandit Problem
Authors: Andrea Locatelli, Maurilio Gutzeit, Alexandra Carpentier
ICML 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | 4. Experiments We illustrate the performance of algorithm APT in a number of experiments. For comparison, we use the following methods which include the state of the art CSAR algorithm of (Chen et al., 2014) and two minor adaptations of known methods that are also suitable for our problem... Figure 1 displays the estimated probability of success on a logarithmic scale with respect to the horizon of the six algorithms based on N = 5000 simulated games with τ = 1 2, ϵ = 0.1, K = 10, and T = 500. |
| Researcher Affiliation | Academia | Andrea Locatelli ANDREA.LOCATELLI@UNI-POTSDAM.DE Maurilio Gutzeit MGUTZEIT@UNI-POTSDAM.DE Alexandra Carpentier CARPENTIER@UNI-POTSDAM.DE Department of Mathematics, University of Potsdam, Germany |
| Pseudocode | Yes | Algorithm 1 APT algorithm Input: τ, ϵ Pull each arm once for t = K + 1 to T do Pull arm It = arg min k K Bk(t) from Equation (5) Observe reward X νIt end for Output: ˆSτ = {k : ˆµk(T) τ} |
| Open Source Code | No | The paper does not contain any explicit statement or link indicating that the source code for the described methodology is publicly available. |
| Open Datasets | No | The experiments are based on 'simulated games' with Bernoulli distributions and specified parameters (e.g., K = 10 arms, means µ1:3 0.1), rather than a publicly available dataset with concrete access information. |
| Dataset Splits | No | The paper mentions 'N = 5000 simulated games' but does not specify any explicit train/validation/test dataset splits. |
| Hardware Specification | No | The paper describes running 'simulated games' but provides no specific details about the hardware used (e.g., GPU/CPU models, memory). |
| Software Dependencies | No | The paper does not specify any software dependencies with version numbers (e.g., Python, PyTorch, specific solvers). |
| Experiment Setup | Yes | Figure 1 displays the estimated probability of success on a logarithmic scale with respect to the horizon of the six algorithms based on N = 5000 simulated games with τ = 1/2, ϵ = 0.1, K = 10, and T = 500. Experiment 1 (3 groups setting): K Bernoulli arms with means µ1:3 0.1, µ4:7 = (0.35, 0.45, 0.55, 0.65) and µ8:10 0.9... |