Learning to Assess the Reliability of Number-of-Runs Estimation in Stochastic Optimization

This repository contains the datasets, extracted features, and experimental results used in the study “Learning to Assess the Reliability of Number-of-Runs Estimation in Stochastic Optimization”.The folder organization is designed to clearly separate raw data, processed features, and experimental outputs, ensuring transparency and reproducibility. Top-level Structure Project- Adaptive estimation uncertainty quantification/│├── datasets/├── features/└── results/ datasets/ This folder contains the original raw data used in the study, provided in two complementary forms: datasets/├── Nevergard_annotation-20251202T093830Z-1-001/│ └── Nevergard_annotation/│├── Nevergard_samples-20251202T093834Z-1-001/│ └── Nevergard_samples/ Nevergard_annotation/ Contains annotation files associated with the dataset. These files define the annotations from the sample data. Nevergard_samples/ Contains the sample data, the number of experimental runs. This is the primary input used for feature extraction and model training. Files are kept in their original format to preserve data integrity. features/ This folder contains the extracted features derived from the samples and annotations datasets: features/└── all_features/ all_features/ Stores the complete set of extracted features used in the experiments. These features serve as the input to all learning models. results/ This directory contains the outputs of all experimental runs reported in the paper: results/├── Experiment_E1/├── Experiment_E2/├── Experiment_E3/├── Experiment_E4/└── Experiment_E5/ Experiment_E1 – Experiment_E5 Each experiment folder corresponds to a distinct experimental configuration as presented in the paper: different uncertainty estimation strategies, different feature subsets, different model settings. This separation ensures that results from different experimental setups do not overlap and can be independently verified.