Approximations to worst-case data dropping: unmasking failure modes
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2024
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866908526009384960 |
|---|---|
| author | Huang, Jenny Y. Burt, David R. Shen, Yunyi Nguyen, Tin D. Broderick, Tamara |
| author_facet | Huang, Jenny Y. Burt, David R. Shen, Yunyi Nguyen, Tin D. Broderick, Tamara |
| contents | A data analyst might worry about generalization if dropping a very small fraction of data points from a study could change its substantive conclusions. Checking this non-robustness directly poses a combinatorial optimization problem and is intractable even for simple models and moderate data sizes. Recently various authors have proposed a diverse set of approximations to detect this non-robustness. In the present work, we show that, even in a setting as simple as ordinary least squares (OLS) linear regression, many of these approximations can fail to detect (true) non-robustness in realistic data arrangements. We focus on OLS in the present work due its widespread use and since some approximations work only for OLS. Across our synthetic and real-world data sets, we find that a simple recursive greedy algorithm is the sole algorithm that does not fail any of our tests and also that it can be orders of magnitude faster to run than some competitors. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2408_09008 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Approximations to worst-case data dropping: unmasking failure modes Huang, Jenny Y. Burt, David R. Shen, Yunyi Nguyen, Tin D. Broderick, Tamara Methodology Computation A data analyst might worry about generalization if dropping a very small fraction of data points from a study could change its substantive conclusions. Checking this non-robustness directly poses a combinatorial optimization problem and is intractable even for simple models and moderate data sizes. Recently various authors have proposed a diverse set of approximations to detect this non-robustness. In the present work, we show that, even in a setting as simple as ordinary least squares (OLS) linear regression, many of these approximations can fail to detect (true) non-robustness in realistic data arrangements. We focus on OLS in the present work due its widespread use and since some approximations work only for OLS. Across our synthetic and real-world data sets, we find that a simple recursive greedy algorithm is the sole algorithm that does not fail any of our tests and also that it can be orders of magnitude faster to run than some competitors. |
| title | Approximations to worst-case data dropping: unmasking failure modes |
| topic | Methodology Computation |
| url | https://arxiv.org/abs/2408.09008 |