Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1]
Improving multiple-try Metropolis with local balancing
Authors: Philippe Gagnon, Florian Maire, Giacomo Zanella
JMLR 2023 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We show both theoretically and empirically that this weight function induces pathological behaviours in high dimensions, especially during the convergence phase. We propose to instead use weight functions akin to the locally-balanced proposal distributions of Zanella (2020), thus yielding MTM algorithms that do not exhibit those pathological behaviours... Numerical experiments include an application in precision medicine involving a computationally-expensive forward model, which makes the use of parallel computing within MTM iterations beneficial. |
| Researcher Affiliation | Academia | Philippe Gagnon EMAIL Department of Mathematics and Statistics Universit e de Montr eal Montreal, Canada Florian Maire EMAIL Department of Mathematics and Statistics Universit e de Montr eal Montreal, Canada Giacomo Zanella EMAIL Department of Decision Sciences and BIDSA Bocconi University Milan, Italy |
| Pseudocode | No | The paper describes the MTM iteration steps in paragraph text (e.g., "In its simplest and most popular form, an iteration of MTM is as follows: 1. N values...") and refers to an algorithm from an external paper ("Algorithm 4 found in Section 5 of Andrieu and Thoms (2008)"), but it does not present any pseudocode or algorithm blocks within its own text. |
| Open Source Code | Yes | The code to reproduce all numerical results is available online.4 (Footnote 4: See ancillary files on ar Xiv:2211.11613.) |
| Open Datasets | Yes | Our data set has been simulated from a virtual cohort (generated as in Jenner et al. (2021)) that is then fed into the model defined in (8). (Jenner et al. (2021) refers to: Adrianne L Jenner, Tyler Cassidy, Katia Belaid, Marie-Claude Bourgeois-Daigneault, and Morgan Craig. In silico trials predict that combination strategies for enhancing vesicular stomatitis oncolytic virus are determined by tumor aggressivity. Journal for immunotherapy of cancer, 9(2):1 13, 2021.) |
| Dataset Splits | No | The paper describes the simulated data set structure: "The simulated data set consists of a virtual cohort of K = 10 patients, where each patient is examined once per week for T = 20 weeks. At each examination, m1 = 4 statistics are measured;". However, it does not specify any training, validation, or test splits, which are typically used for evaluating model performance in many experimental setups. For Bayesian inference, the entire dataset is typically used for parameter estimation, not split for generalization testing in the traditional sense. |
| Hardware Specification | Yes | In our experiment, we used a desktop computer with 32 cores (thus larger than the number of candidates N), AMD Ryzen 9 5950X processor, Alma Linux 8.5 operating system, 64 GB of RAM, and off-the-shelf high-level MATLAB parallelization. |
| Software Dependencies | No | The paper mentions "Alma Linux 8.5 operating system" and "off-the-shelf high-level MATLAB parallelization". While Alma Linux includes a version, MATLAB is mentioned without a specific version number. No other software or library versions are provided. |
| Experiment Setup | Yes | Algorithm 4 found in Section 5 of Andrieu and Thoms (2008) is used to adaptively tune σ. The algorithm targets an acceptance rate to adapt tuning parameters. The targeted acceptance rates are 25% and 50% for GB and LB MTM, respectively... Algorithm 4 of Andrieu and Thoms (2008) also uses a learning rate γ(m), which here is set to m 0.6, m representing the iteration index... All samplers use the proposal qσ(x, ) = N(x, σ2Id), with σ = ℓ/d. |