Summary of Tabred: Analyzing Pitfalls and Filling the Gaps in Tabular Deep Learning Benchmarks, by Ivan Rubachev et al.
TabReD: Analyzing Pitfalls and Filling the Gaps in Tabular Deep Learning Benchmarks
by Ivan Rubachev, Nikolay Kartashev, Yury Gorishniy, Artem Babenko
First submitted to arxiv on: 27 Jun 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: None
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The paper explores the gap between machine learning research and its practical application in real-world scenarios. It highlights two key characteristics of tabular data in industrial settings that are underrepresented in academic evaluation datasets: temporal distribution drift and feature engineering pipelines. The authors analyze existing benchmarks and find that time-based train/test splits are necessary to account for distribution drift, but most popular datasets lack timestamp metadata. They also show that extensive feature engineering can lead to more predictive features, uninformative features, and correlations. To bridge the gap between research and industry, the authors introduce TabReD, a collection of eight industry-grade tabular datasets, and reassess various tabular ML models on these new benchmarks. The results demonstrate that evaluation on time-based data splits leads to different method rankings compared to random splits. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper looks at how machine learning research can be used in real-life situations. It finds two things that are often missing from the datasets we use to test our methods: changes over time and extra work done to prepare the data. The authors think that if we account for these changes, we might get different results than we would with the current ways of testing. To help fix this problem, they created a new set of datasets that are more like what you’d find in industry. They tested some popular machine learning methods on these new datasets and found that some worked better than others. |
Keywords
» Artificial intelligence » Feature engineering » Machine learning
