How mit-mi/smellnet works
SmellNet is a sensor-based smell-recognition benchmark and research pipeline. Its main workflow runs from MOX sensor recordings through preprocessing and sliding-window preparation into single-substance or mixture modeling, evaluation, and predictions. Separate branches support serial data capture, GC-MS spectrum processing for optional training-time sensor-chemistry alignment, and exploratory analyses. The README documents the overall workflow; source-level evidence is limited to the sampled files and SOURCE INDEX, so uncertain implementation links are omitted.
Main entry points: run.py Base experiments · run_mixture.py Mixture experiments · dataset.py Windowed datasets · load_data.py Sensor data loading
Components 19
Sensing and acquisition
- MOX sensors
- Sensor drivers
Arduino/libraries/Adafruit_BME680_Library/bme68x.c - Serial capture
data_collection/serial-data-collect-csv.py
Data preparation
- Raw-data processing
preprocessing/process_raw_data.py - Column correction
preprocessing/correct_column.py - Sensor data loading
models/load_data.py - Windowed datasets
models/dataset.py
Modeling and evaluation
- Base experiments
models/run.py - Mixture experiments
models/run_mixture.py - Temporal architectures
models/models.py - Training and losses
models/train.py - Evaluation
models/evaluate.py - Recognition results
GC-MS supervision
- Food identifiers
gcms_analysis/food.py - GC-MS processing
gcms_analysis/gcms_analysis.py - GC-MS analysis
gcms_analysis/analysis.py - GC-MS vectors
Analysis and figures
- Sensor analyses
analysis/tsne_analysis.py - GC-MS figures
gcms_analysis/gcms_paper_figures.py
Connections 21
- MOX sensors → Sensor drivers: measured by
- MOX sensors → Serial capture: recorded by
README.md - Serial capture → Raw-data processing: supplies recordings
- Raw-data processing → Column correction: prepares data
- Sensor data loading → Windowed datasets: segments windows
README.md - Windowed datasets → Base experiments: feeds training
README.md - Windowed datasets → Mixture experiments: feeds training
README.md - Base experiments → Temporal architectures: builds models
models/run.py - Mixture experiments → Temporal architectures: builds models
models/run_mixture.py - Base experiments → Training and losses: trains
models/run.py - Base experiments → Evaluation: evaluates
models/run.py - Base experiments → Recognition results: produces
README.md - Mixture experiments → Training and losses: trains
- Mixture experiments → Recognition results: produces ratios
README.md - Food identifiers → GC-MS analysis: supplies food IDs
gcms_analysis/analysis.py - Food identifiers → GC-MS processing: supplies food IDs
gcms_analysis/gcms_analysis.py - GC-MS processing → GC-MS vectors: generates vectors
README.md - GC-MS analysis → GC-MS vectors: generates vectors
README.md - GC-MS vectors → Base experiments: optional supervision
README.md - Sensor data loading → Sensor analyses: analyzes signals
- GC-MS vectors → GC-MS figures: visualized by