Datasets:

HTW-KI-Werkstatt
/

RamanBench

The number of columns (11086) exceeds the maximum supported number of columns (1000). This is a current limitation of the datasets viewer. You can reduce the number of columns if you want the viewer to work.

Error code:   TooManyColumnsError

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

RamanBench Dataset Mirror ⚠️ RESEARCH MIRROR ONLY

⚠️ IMPORTANT: This is a research/benchmarking mirror only.

All datasets are provided for research and educational purposes. Original copyrights, ownership, and intellectual property rights remain with the original dataset authors and institutions. See Original Dataset Sources & Licenses below for details.

A unified mirror of all Raman spectroscopy datasets from the RamanBench benchmark on HuggingFace Hub.

Overview

This dataset repository consolidates 84+ Raman spectroscopy datasets from diverse sources (Kaggle, Zenodo, Figshare, GitHub, etc.) into a single, unified interface for research purposes only. Each dataset is available as a separate configuration, enabling fast, reliable access without depending on external sources.

Purpose: Accelerate research in Raman spectroscopy by providing:

✅ Fast, reliable dataset downloads (HF Hub CDN)
✅ Consistent wide-format Parquet storage (wavenumber columns + targets)
✅ Complete metadata (task type, target names)
✅ Support for both classification and regression tasks
✅ Multi-target (multi-analyte) dataset handling

Citation & Attribution

If using this mirror:

Please cite the RamanBench paper:

@article{Koddenbrock2026RamanBench,
  title={RamanBench: A Comprehensive Benchmark for Machine Learning on Raman Spectroscopy},
  author={Koddenbrock, Mario and Mahrova, Marija and Schulz, Alexander and Gruen, Dominic and others},
  journal={arXiv preprint arXiv:2605.02003},
  year={2026}
}

CRITICAL: Original Dataset Attribution

When using any dataset from this mirror, you MUST ALSO cite the original source. Each dataset has original authors, licenses, and terms of use. Consult the Original Dataset Sources & Licenses table below and the original publications/repositories.

Failure to cite original sources violates intellectual property rights.

Usage

Load any dataset with HuggingFace's load_dataset:

from datasets import load_dataset

# Load a single dataset
ds = load_dataset("HTW-KI-Werkstatt/RamanBench", "wheat_lines", split="train")
print(ds.features)  # All wavenumber columns + target

# Access data
df = ds.to_pandas()
wavenumbers = df[[col for col in df.columns if col != 'target']]
targets = df['target']

# Load multi-target dataset (e.g., bioprocess analytes)
ds = load_dataset("HTW-KI-Werkstatt/RamanBench", "bioprocess_analytes_anton_532", split="train")
# Columns: wavenumber floats + analyte targets (glucose, ethanol, etc.)

Dataset Structure

Each configuration (dataset) contains:

Parquet files: Wide-format spectral data
- Columns: Wavenumber values as column names (e.g., "200.5", "250.0") + target(s)
- Data types: float32 for spectra, float64 or string for targets
- Rows: Individual spectrum samples
Metadata: Target names and task type (classification/regression)

Example: Single-target dataset

wheat_lines/
├── train-00000-of-00001.parquet
│   ├── 99.9, 100.2, 100.5, ..., 898.8 (wavenumber columns, float32)
│   ├── target (classification labels as strings)
└── metadata.json
    {"target_names": ["target"], "task_type": "classification"}

Example: Multi-target dataset

bioprocess_analytes_anton_532/
├── train-00000-of-00001.parquet
│   ├── 99.9, 100.2, 100.5, ..., 898.8 (wavenumber columns, float32)
│   ├── glucose (float64)
│   ├── ethanol (float64)
│   ├── glycerol (float64)
│   └── acetic_acid (float64)
└── metadata.json
    {"target_names": ["glucose", "ethanol", "glycerol", "acetic_acid"], "task_type": "regression"}

Available Datasets

84 datasets across classification and regression tasks:

Classification (e.g., material/species identification)

wheat_lines — Wheat varieties
bacteria_identification — Bacterial species
hair_dyes_sers — Hair dye compounds (SERS)
cancer_cell_cooh, cancer_cell_nh2, cancer_cell_(cooh)2 — Cancer cell detection
And many more...

Regression (e.g., concentration prediction, analyte quantification)

amino_acids_* (glycine, leucine, phenylalanine, tryptophan) — Amino acid concentrations
bioprocess_analytes_* (anton_532, anton_785, kaiser, metrohm, mettler_toledo, tec5, timegate, tornado) — Fermentation monitoring (multi-target)
bioprocess_substrates — Substrate concentrations (multi-target)
ecoli_fermentation — E. coli fermentation tracking
And many more...

See the RamanBench paper for complete dataset descriptions.

⚠️ Original Dataset Sources & Licenses

COPYRIGHT & IP RIGHTS: Each dataset retains its original copyright and intellectual property ownership. The authors, institutions, and original repositories listed below own all rights to their data. This mirror is for research access only.

USAGE REQUIREMENTS:

✅ You MAY use these datasets for research and non-commercial purposes
✅ You MUST cite the original authors and sources
✅ You MUST respect the original dataset licenses (see below)
❌ You CANNOT redistribute datasets without original author permission
❌ You CANNOT use for commercial purposes unless original licenses permit

Dataset Sources & Licenses

Source Repository	Datasets	Original License	Copyright Holder	Citation
Kaggle	bioprocess_analytes_* (multiple instruments), ecoli_, fuel_	Dataset Terms	Dataset authors/Kaggle	Kaggle dataset pages
Zenodo	rruff_mineral_, microgel_, organic_compounds_*	Varies (CC-BY, CC-0, custom)	Original depositors	Zenodo repository DOIs
Figshare	cancer_cell_, diabetes_skin_, serum_alzheimer, serum_prostate	CC-BY (default)	Original depositors	Figshare DOIs
GitHub (HTW-KI-Werkstatt)	fuel_handheld, ecoli_metabolites_dig4bio, yeast_fermentation, ralstonia_fermentations	CC-BY or dataset-specific	HTW-KI-Werkstatt / Contributors	GitHub repositories
HuggingFace Hub	wheat_lines, bacteria_identification, and others	Varies per dataset	Original contributors	HuggingFace dataset cards
RWTH Aachen / Custom	Various research datasets	Institution-dependent	Original research groups	Publication/repository links in dataset cards

Before Using Any Dataset:

Check the original source for the most current license and terms
Cite the original authors — not just RamanBench
Respect usage restrictions — some datasets may have commercial or redistribution limitations
Contact original authors if unclear about usage rights

This mirror is a convenience for research — NOT a transfer of intellectual property or rights.

Use in RamanBench

This mirror is integrated into the RamanBench package:

from raman_bench import RamanBenchmark

# Load benchmark using this mirror (default)
bench = RamanBenchmark(
    dataset_names_classification=["wheat_lines"],
    dataset_names_regression=["amino_acids_glycine"],
    use_mirror=True  # Uses HF mirror (fast)
)

# Or load from original sources
bench_original = RamanBenchmark(
    dataset_names_classification=["wheat_lines"],
    dataset_names_regression=["amino_acids_glycine"],
    use_mirror=False  # Uses original raman_data loaders
)

Metadata Format

Each dataset includes a metadata.json file:

{
  "target_names": ["target"],
  "task_type": "classification"
}

or (multi-target):

{
  "target_names": ["glucose", "ethanol", "glycerol", "acetic_acid"],
  "task_type": "regression"
}

⚠️ Disclaimer & Legal Notice

MIRROR STATUS:

This is a research-focused mirror for benchmarking and ML experiments ONLY
NOT authorized for commercial use without explicit permission from original authors
Availability depends on HuggingFace Hub uptime — original sources are the primary repositories

INTELLECTUAL PROPERTY & COPYRIGHT: