GRAND-SLAMIN’ Interpretable Additive Modeling with Structural Constraints
Authors: Shibal Ibrahim, Gabriel Afriat, Kayhan Behdin, Rahul Mazumder
NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Numerical experiments on real-world datasets show that our toolkit performs favorably in terms of performance, variable selection and scalability when compared with popular toolkits to fit GAMs with interactions. |
| Researcher Affiliation | Academia | Shibal Ibrahim MIT Cambridge, MA shibal@mit.edu; Gabriel Isaac Afriat MIT Cambridge, MA afriatg@mit.edu; Kayhan Behdin MIT Cambridge, MA behdink@mit.edu; Rahul Mazumder MIT Cambridge, MA rahulmaz@mit.edu |
| Pseudocode | No | The paper describes methodologies but does not include a clearly labeled pseudocode block or algorithm section. |
| Open Source Code | Yes | Our code is available at https://github.com/mazumder-lab/grandslamin. |
| Open Datasets | Yes | We use a collection of 16 open-source classification datasets (binary, multiclass and regression) from various domains... They are from Penn Machine Learning Benchmarks (PMLB) [41] and UCI databases [8]. |
| Dataset Splits | Yes | When no test set was available, we treated the original validation set as the test set and split the training set into 80% training and 20% validation. For remaining, we randomly split each of the dataset into 60% training, 20% validation and 20% testing sets. |
| Hardware Specification | Yes | We used a cluster running Ubuntu 7.5.0 and equipped with Intel Xeon Platinum 8260 CPUs and Nvidia Volta V100 GPUs. For all experiments of Sec. 6, each job involving GRAND-SLAMIN, EBM, Node-GAM, SIAN, GAMI-Net and DNN were run on 8 core, 32GB RAM. Jobs involving larger datasets (p > 100) were run on Tesla V100 GPUs. |
| Software Dependencies | No | The paper mentions Ubuntu 7.5.0 (an operating system) and Optuna [2] for hyperparameter tuning but does not specify version numbers for other core software dependencies (e.g., programming languages, deep learning frameworks, or libraries) required for replicating the methodology. |
| Experiment Setup | Yes | Learning Rates: Discrete uniform in the set {0.05, 0.01, 0.005} for Adam with multi-step decay rate of 0.9 every 25 epochs. Batch-size: Discrete uniform in the set {64, 256}. λ for selection: Discrete uniform in the set of 11 values {0, 1e 6, , 1e 3}. γ for Smooth-step: Discrete uniform in the set {0.01, 0.1, 1}. τ for Entropy Regularization: Discrete uniform in the set {0.001, 0.01, 0.1}. α for relative penalty on interactions: Discrete uniform in the set {1, 10}. Epochs: 1000 with early stopping (patience=50) based on validation loss. |