smartnanotubes/snt-fire-enose-sample

SNT Fire/Smoke E-Nose - Sample Data Subset

A small, curated sample of raw recordings from SmartNanotubes' 4×16-channel carbon-nanotube (CNT) electronic-nose arrays, released so that researchers and partners can see the signal quality and experiment with the data. This is a teaser subset, not the full training corpus.

It accompanies the model card at smartnanotubes/snt-fire-enose-5class and the interactive demo at smartnanotubes/snt-fire-enose-demo.

What's in it

Each recording is a time series from the CNT array sampled at 1 Hz, labeled with one of five air-quality classes:

1. Normal air (background)
2. Cigarette smoke
3. Cooking / frying aerosols
4. Smoldering electrical cable
5. Smoldering fabric / tissue

• Recordings: 14 sessions across 10 devices and 6 sites (~10.9 hours total at 1 Hz)
• Device: each device carries four 16-channel SNT sensor chips plus a temperature and a relative-humidity sensor
• Columns: time_s, ch_00 … ch_63 (raw resistance values per active channel, ordered as four 16-channel arrays), temperature, humidity, class label
• Format: one CSV per recording under data/; see manifest.csv for per-recording device/site tokens and per-class row counts
• Class balance: normal-air dominates (~70%), as in real deployment; the four hazard classes are the minority - use macro-averaged metrics, not accuracy

What is not provided (privacy / IP)

To protect proprietary information, the following are deliberately stripped and are not present in this subset:

• Device serial numbers, hardware batch identifiers, and site/location names
• Absolute timestamps and calendar dates
• Any details on smell-sensor chip composition
• The engineered features, preprocessing, and postprocessing used by SNT's original demo model

Raw channel readings (resistance values in Ohm) are provided as-is. The proprietary value is in the feature engineering and models built on top of this signal - which are not part of this release.

Intended use

• Explore CNT e-nose signal characteristics.
• Prototype your own classifiers and share results.
• Reproduce the baseline used in the demo Space.

Disclaimer: this subset is not sufficient to train a production-grade ML model, is not a certified dataset for safety applications, and carries no concentration/ppm labels.

How to load

from datasets import load_dataset

ds = load_dataset("smartnanotubes/snt-fire-enose-sample")
print(ds)

License & contact

Released under a proprietary sample-data license (see LICENSE). For the full dataset, evaluation access, or partnership, contact SmartNanotubes Technologies - info@smart-nanotubes.com