Skip to content

Reporting & Reproducibility Guidelines

Purpose: Document your analysis results reproducibly and support peer review.
Audience: Researchers publishing FoodSpec workflows or sharing results.
Time to read: 8–12 minutes.
Prerequisites: Completed a FoodSpec workflow run; familiar with classification/regression metrics.

What to present as core results

  • Main figures: confusion matrix (classification), PCA scores, key ratio/time trends, predicted vs true plots (mixture/regression)
  • Main tables: overall accuracy/F1 (classification) or R²/RMSE (regression/mixture); include fold-averaged metrics
  • Preprocessing summary: list methods/parameters (baseline, smoothing, normalization, cropping)
  • Model and validation: classifier/regressor type, CV design (folds, stratification), seeds

What belongs in supplementary material

  • Full per-class precision/recall/F1 tables, additional confusion matrices
  • Hyperparameters, alternative models tried, sensitivity analyses
  • Additional spectra/ratios, extended residual plots, run_metadata.json/config files

Describing methods for reproducibility

  • Data origin: State data origin, modality, instrument, sample prep conditions
  • Preprocessing: Describe preprocessing steps with parameters; note wavenumber range
  • Features: Specify features used (peaks/ratios/bands) and model choices
  • Validation: Document validation setup: CV splits, stratification, metrics reported, any held-out test sets
  • Reproducibility: Reference CLI/Python commands or configs used (e.g., foodspec oil-auth with flags, or script snippets)

Follow-up & supporting tests

  • External validation: Independent dataset or instrument for validation
  • Orthogonal analyses: E.g., GC–MS, peroxide/anisidine values, sensory tests to corroborate spectroscopy results
  • Robustness checks: New batches, different preprocessing, or small perturbations to confirm stability

FAIR data principles

Align with FAIR (Findable, Accessible, Interoperable, Reusable): - Keep data + metadata together - Cite public datasets/DOIs - Share configs and run artifacts when possible - Document computational environment (Python version, package versions)

Next Steps