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 Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Learning Gradient Boosted Decision Trees with Algorithmic Recourse
Authors: Kentaro Kanamori, Ken Kobayashi, Takuya Takagi
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conducted numerical experiments on real datasets and demonstrated that our RABIT successfully provided reasonable actions to more instances than the baselines without significantly degrading accuracy and computational efficiency. |
| Researcher Affiliation | Collaboration | Kentaro Kanamori Artificial Intelligence Laboratory Fujitsu Limited EMAIL Ken Kobayashi School of Engineering Institute of Science Tokyo EMAIL Takuya Takagi Artificial Intelligence Laboratory Fujitsu Limited EMAIL |
| Pseudocode | Yes | Algorithm 1 Algorithm for approximately solving the problem (5). Algorithm 2 Actionable feature tweaking algorithm for approximately solving the problem (1). |
| Open Source Code | Yes | All the code was implemented in Python 3.10 with Numba 0.61.0 and is available at https://github.com/kelicht/rabit. |
| Open Datasets | Yes | We used four real benchmark datasets: FICO (N = 9871, D = 23) [14], COMPAS (N = 6167, D = 14) [2], Adult (N = 48842, D = 16) [34], and Bail (N = 8923, D = 16) [53]. All the datasets used in Section 5 are publicly available and do not contain any identifiable information or offensive content. |
| Dataset Splits | Yes | We randomly split the dataset into the training and test sets with a ratio of 75 : 25, and trained tree ensemble models by each method on the training set. We randomly split the dataset into the training, calibration, and test sets with a ratio of 50 : 25 : 25. |
| Hardware Specification | Yes | All the experiments were conducted on mac OS Sequoia with Apple M2 Ultra CPU and 128 GB memory. |
| Software Dependencies | Yes | All the code was implemented in Python 3.10 with Numba 0.61.0 and is available at https://github.com/kelicht/rabit. Numba 0.61.0 is publicly available under the BSD-2-Clause license. Py SCIPOpt 5.5.0 is publicly available under the MIT license. |
| Experiment Setup | Yes | For the baselines and our RABIT, we trained T = 100 regression trees with a maximum depth of 8 and a learning rate 0.1. For RABIT, we set γ = 0.002 and did not apply our leaf refinement post-processing. |