Human Guided Linear Regression With Feature-Level Constraints
Authors: Aubrey Gress, Ian Davidson
AAAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our experimental results show our formulations outperform natural baselines and prior work (Section 5). ... The methods and data sets we used are described in Tables 1 and 2. We generated the guidance by simulating a domain expert. ... Reported results are the mean of 30 train/test splits. Values in parentheses indicate 95% confidence interval. More extensive experiments and learning curves are presented in the supplementary material. |
| Researcher Affiliation | Academia | Aubrey Gress, Ian Davidson Department of Computer Science, University of California, Davis 1 Shields Avenue, Davis, CA, 95616 {adgress@ucdavis.edu, davidson@cs.ucdavis.edu} |
| Pseudocode | No | The paper describes methods using mathematical formulations and descriptive text, but does not include any explicitly labeled 'Pseudocode' or 'Algorithm' blocks. |
| Open Source Code | Yes | Code and processed data sets are available at https://github.com/adgress/AAAI2018. |
| Open Datasets | Yes | Boston Housing (Harrison and Rubinfeld 1978; Lichman 2013) ... Wine (Lichman 2013) The UCI wine data set. ... Concrete (Yeh 1998; Lichman 2013) ... Lichman, M. 2013. UCI machine learning repository. |
| Dataset Splits | Yes | Reported results are the mean of 30 train/test splits. Values in parentheses indicate 95% confidence interval. ... For all methods all regularization parameters were tuned on a validation set. |
| Hardware Specification | No | The paper does not provide specific details regarding the hardware used for running the experiments (e.g., GPU/CPU models, memory specifications). |
| Software Dependencies | No | The paper mentions 'Sci Py s optimization module (Jones et al. 2001)' but does not provide specific version numbers for any software dependencies required to replicate the experiments. |
| Experiment Setup | Yes | For all methods all regularization parameters were tuned on a validation set. ... Specifically, we randomly selected a subset of features, estimate the signs (or relative orderings) of regression coefficients of each feature separately using ordinary least squares on a validation set. i.e. to generate the sign of coefficient i, we solved the problem ˆβi = arg minβi ||Xiβi Y ||2 and used the sign to generate the guidance. |