How visserle/pain-measurement works
Project type and purpose
• This is a research-oriented data pipeline combining experimental data collection (physiological signals + questionnaires), data ingestion/storage, feature engineering, ML model training/evaluation, and reporting/visualization.
• Key use cases: run calibration & measurement sessions (audio prompts, hardware integration, Imotions), collect questionnaire responses via a small web app, ingest & anonymize raw data into a database, extract multi-modal features, train and compare time-series and ensemble models, and generate plots/notebooks for analysis.
High-level architectural layers
A. Data Collection Layer
– Calibration (src/experiments/calibration): scripts, audio files, estimator.
– Measurement (src/experiments/measurement): stimulus generator, rate limiter, hardware interface (imotions.py, thermoino.py), audio prompts, config/scripts.
– Questionnaires (src/experiments/questionnaires): Flask-like app.py, Jinja2 templates, CSS/static assets, scoring_schemas.
B. Data Ingestion & Storage Layer
– raw data → src/data: Imotions importer, anonymize.py, data_processing.py, quality_checks.py
– database_manager.py & database_schema.py hold data in local/remote DB
– YAML/TOML config files drive all ingestion steps
C. Feature Engineering Layer
– src/features: per-modality modules (eda.py, eeg.py, hr.py, pupil.py, face.py) plus filtering, resampling, scaling, transforming, label generation
– exploratory scripts under src/features/exploratory for EDA
D. ML Pipeline Layer
– src/models:
• data_preparation.py & sample_creation.py prepare train/test splits
• data_loader.py & scalers.py standardize inputs
• architectures (src/models/architectures): EEGNet, PatchTST, TimesNet, LSTM, ensembles, transformer blocks
• training_loop.py, hyperparameter_tuning.py, model_selection.py
• evaluation.py records metrics → results JSON, picks best checkpoints
– main.py + main_config.py drive end-to-end training runs
E. Reporting & Visualization Layer
– src/plots: scripts to generate figures (model performance, correlations, calibration curves, stimulus plots)
– Jupyter notebooks for interactive EDA & result presentation (notebooks/*.ipynb)
– reports/figures for static assets to include in write-ups
F. Orchestration & Results Storage
– run_experiments.ps1, run_model_comparison.ps1: wrap and schedule experiments
– results/ folder: hierarchies per experiment with model checkpoints, results.json, summary.txt
Relationships & data flow
1. Participant session → Calibration & Measurement modules (audio, hardware, questionnaire app)
→ raw data files
2. Raw data files → Data Ingestion layer → Database
3. DB records → Feature Engineering layer → Feature datasets
4. Feature datasets → ML Pipeline → Trained models + metric outputs (results/)
5. Results + models → Reporting layer (plots & notebooks) → reports
Architectural patterns & design principles
• Layered decomposition (collection → ingestion → features → modeling → reporting)
• Modular subpackages (clear Python modules for each concern)
• Config-driven behavior (YAML/TOML + pyproject)
• Reusable components (utilities in src/experiments/utils.py, src/features/utils.py, src/models/utils.py)
• Experiment reproducibility (scripts + versioned result directories)
Final pointers
– Be explicit about storage locations (DB vs. results folder).