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 and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
Learning Optimal Group-structured Individualized Treatment Rules with Many Treatments
Authors: Haixu Ma, Donglin Zeng, Yufeng Liu
JMLR 2023 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive empirical results in simulation studies and real data analysis demonstrate that GROWL enjoys better performance than several other existing methods. |
| Researcher Affiliation | Academia | Haixu Ma EMAIL Department of Statistics and Operations Research University of North Carolina at Chapel Hill Chapel Hill, NC 27599, USA; Donglin Zeng EMAIL Department of Biostatistics University of North Carolina at Chapel Hill Chapel Hill, NC 27599, USA; Yufeng Liu EMAIL Department of Statistics and Operations Research Department of Genetics Department of Biostatistics Carolina Center for Genome Science Lineberger Comprehensive Cancer Center University of North Carolina at Chapel Hill Chapel Hill, NC 27599, USA |
| Pseudocode | Yes | Appendix E. Implementation Details: Algorithm 1: GROWL using the Genetic Algorithm; Algorithm 2: GROWL with the Greedy Adjustment |
| Open Source Code | No | The paper mentions using "the R package called GA introduced in Scrucca (2013)" but does not explicitly state that the authors' own implementation code for GROWL is open-source, nor does it provide a direct link to such code. The license information provided is for the paper itself, not for any associated code. |
| Open Datasets | Yes | In this section, we apply our proposed GROWL to analyze the data from the STAR*D study (Rush et al., 2004). |
| Dataset Splits | Yes | We first randomly split the observed data tp Xi, Ai, Riqun i 1 into two folds. For each group number 1 ď K ď Mn, denote the pδK and p Dg,K as the estimated optimal partition and associated group-based decision rule learned from one fold of the training data based on the implementations discussed in Section 2.3. Then, we calculate the estimated value function p V1ppδK, p Dg,Kq for each K using ... Comparisons of all these methods were based on 200 repetitions of three-fold cross-validation, where two folds are used to train the model. |
| Hardware Specification | No | The paper does not provide specific details about the hardware (e.g., CPU, GPU models, memory) used to run the experiments. It focuses on the methodology and experimental results without specifying the computational environment. |
| Software Dependencies | Yes | We formulate the partition space as the discrete problem of partitioning Mn numbers into Kn groups. To solve this non-convex integer programming problem, when Mn and Kn are relatively small, we can implement the genetics algorithm using the R package called GA introduced in Scrucca (2013). |
| Experiment Setup | Yes | The tuning parameter λn is chosen to maximize the empirical value function En RIr p Dp Xq As{pp A|Xq L En Ir p Dp Xq As{pp A|Xq by 10-fold cross-validation among t 1 2, 1, 2, 4, 8, 16u. For the Gaussian kernel, we fix the inverse bandwidth of the kernel σn with 1{p2pτ 2q, where pτ is the median of the pairwise Euclidean distance of the covariates (Wu and Liu, 2007). The treatment group number is determined by the trade-off procedure (14) with T 50 shown in Section 2.4. |