repo-pretrained-dataset-inference

To create a PMLB inference script, you place a new Python file in the experiments/pmlb directory that reads data from pmlb/datasets, uses recommended stronger hyperparameters, and outputs results to experiments/pmlb/results. This extends your existing verified pretrained repository example into a standardized evaluation workflow.

The standard directory structure for PMLB evaluation requires placing inference scripts in experiments/pmlb and storing evaluation results in experiments/pmlb/results. This strict placement ensures outputs are organized and reproducible when extending pretrained repository examples for new experiments.

To update RUN.md for a PMLB inference experiment, append the directly reproducible execution command after the existing pretrained example command, ensuring uniform experiments/pmlb path formatting. This maintains a sequential and standardized command history for project reproducibility.

Yes, you can inject multiplicative Gaussian noise into target variables (y) during PMLB dataset inference using a specified format. This feature allows controlled dataset sampling and robust evaluation while maintaining strict metric and output ordering for the experiment results.

Yes, running deep neural network inference on PMLB datasets requires GPU-first execution without silent degradation. The workflow operates using a uv-based Python runtime to ensure experiments with deep neural networks execute reliably and maintain strict metric output ordering.

A limitation of default PMLB dataset sampling is that it only takes the first 200 rows for rapid validation. While this provides quick experiment verification, it restricts the evaluation scope and requires adjusting project-recommended stronger hyperparameters for full dataset assessments.