Trompt: Towards a Better Deep Neural Network for Tabular Data
Authors: Kuan-Yu Chen, Ping-Han Chiang, Hsin-Rung Chou, Ting-Wei Chen, Tien-Hao Chang
ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The experimental results demonstrate that Trompt outperforms stateof-the-art deep neural networks and is comparable to tree-based models (Figure 1). The performance of Trompt is evaluated on the Grinsztajn45 benchmark and compared with three deep learning models and five tree-based models. The experiments are conducted on a recognized tabular benchmark, Grinsztajn45. Additionally, we add two well-performed tree-based models, Light GBM (Ke et al., 2017) and Cat Boost (Prokhorenkova et al., 2018) to baselines. Thorough empirical studies and ablation tests were conducted to verify the design of Trompt. |
| Researcher Affiliation | Collaboration | 1Sino Pac Holdings, Taipei, Taiwan 2Department of Electronic Engineering, National Cheng Kung University, Tainan, Taiwan. |
| Pseudocode | No | The paper describes the architecture and processes using diagrams and equations, but it does not include any clearly labeled pseudocode or algorithm blocks. |
| Open Source Code | No | The paper links to the Grinsztajn45 benchmark’s GitHub repository, but it does not state that the source code for Trompt (their proposed method) is open-source or provide a link for it. For example, it does not say “We release our code” or “Our source code is available at…” |
| Open Datasets | Yes | The performance and ablation study of Trompt primarily focus on the Grinsztajn45 benchmark (Grinsztajn et al., 2022). Grinsztajn45 (Grinsztajn et al., 2022) selects 45 tabular datasets from various domains mainly provided by Open ML (Vanschoren et al., 2013) |
| Dataset Splits | No | The paper mentions following “train test data split” configurations from Grinsztajn45 but does not explicitly detail a validation split or its percentages within this paper. While it states that performance was evaluated using the “lowest validation loss” for some datasets, it does not provide specific information on how the validation splits were created or their sizes, referring back to the Grinsztajn45 paper for general configurations without quoting explicit split details. |
| Hardware Specification | No | The paper does not provide specific details about the hardware used for running experiments, such as GPU models, CPU types, or memory specifications. It does not mention “NVIDIA A100”, “Intel Xeon”, or similar specific hardware. |
| Software Dependencies | No | The paper states that “Trompt is implemented using Py Torch,” but it does not specify version numbers for PyTorch or any other software dependencies (e.g., Python, CUDA, other libraries). |
| Experiment Setup | Yes | The default hyperparameters are shown in Table 2. (Table 2 lists parameters like Prompts (P=128) and Cells (L=6)). Please refer to Appendix F for the hyperparameter search spaces for all baselines and Trompt. |