Optimal Thompson Sampling strategies for support-aware CVaR bandits
Authors: Dorian Baudry, Romain Gautron, Emilie Kaufmann, Odalric Maillard
ICML 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Further, we illustrate empirically the benefit of Thompson Sampling approaches both in a realistic environment simulating a use-case in agriculture and on various synthetic examples. ... 4. Experiments |
| Researcher Affiliation | Academia | 1Univ. Lille, CNRS, Inria, Centrale Lille, UMR 9198-CRISt AL, F-59000 Lille, France 2CIRAD, UPR AIDA, F-34398 Montpellier, France 3CGIAR Platform for Big Data in Agriculture, Alliance of CIAT and Bioversity International, Km 17 Recta Cali-Palmira, Apartado Aéreo 6713, Cali, Colombia. |
| Pseudocode | Yes | Algorithm 1 M-CVTS ... Algorithm 2 B-CVTS |
| Open Source Code | No | For the sake of the experiments, we built a bandit-oriented Python wrapper to DSSAT that we made available4 to the bandit community for reproducibility. 4 https://github.com/rgautron/Dssat Bandit Env. While this provides code for their experimental setup, it is not explicitly stated to be the source code for the core methodologies (M-CVTS and B-CVTS) themselves. |
| Open Datasets | Yes | We specifically address maize planting date decision, as maize is a crucial crop for global food security (Shiferaw et al., 2011). Each simulation is assumed to be realistic, and starts from the same field initial conditions as ground measured. The simulator takes as input historical weather data, field soil measures, crop specific genetic parameters and a given crop management plan. ... DSSAT is an Open-Source project maintained by the DSSAT Foundation, see https://dssat.net/. |
| Dataset Splits | No | The paper does not provide specific training, validation, or test dataset splits in the traditional sense, as it focuses on sequential decision-making in a bandit setting or simulated data generation rather than static dataset partitioning. |
| Hardware Specification | No | Experiments presented in this paper were carried out using the Grid 5000 testbed, supported by a scientific interest group hosted by Inria and including CNRS, RENATER and several Universities as well as other organizations (see https://www.grid5000.fr). While a testbed is mentioned, no specific hardware components like GPU/CPU models or memory details are provided. |
| Software Dependencies | No | For the sake of the experiments, we built a bandit-oriented Python wrapper to DSSAT that we made available. ... DSSAT, standing for Decision Support System for Agrotechnology Transfer... The paper mentions Python and DSSAT but does not provide specific version numbers for Python or any key libraries, only DSSAT3 for the simulator itself. |
| Experiment Setup | Yes | We tested the TS algorithms on specified difficult instances and on randomly generated problems, against U-UCB and CVa R-UCB. ... with 103 random instances with 5 arms... for α {10%, 50%, 90%} and an horizon 104. ... averaged over 400 random instances with K = 30, α = 5% (results: mean (std)). ... Experiments are carried out with an horizon of 104 time steps, and we compare the results for each algorithm for α {5%, 20%, 80%} |