DEAP Emotion Recognition Workspace

Research code for EEG-based emotion recognition on the DEAP dataset. The repository combines a configurable classical machine-learning pipeline with several deep-learning experiment branches, evaluation scripts, and report assets used in the graduation project.

What Is In The Repo

• src/deap_emotion/: preprocessing, feature extraction, labeling, metrics, and experiment orchestration
• run_pipeline.py: quick baseline evaluation for subject-dependent and cross-subject runs
• run_experiments.py: configurable experiment runner with model search and report generation
• eeg_*: model-specific deep-learning experiments such as EEGNet, EEGNet+Transformer, ConvNeXt-EEG, TSception, and DE-based variants
• reports/: generated summaries, paper notes, and diagram sources
• tests/: smoke and unit tests for the core pipeline

Quick Start

Install the root dependencies:

pip install -r requirements.txt

Place the DEAP subject .dat files in archive/.

Run the baseline pipeline:

python run_pipeline.py --mode both --output reports/metrics.json

Run a configurable experiment sweep:

python run_experiments.py --models xgb rbf_svm --outer-repeats 5 --inner-folds 4 --scoring f1_macro

Run the test suite:

pytest -q

Notes

• The main pipeline supports multiple label schemes, configurable feature sets, feature selection, and trial-level aggregation.
• Deep-learning folders ship with their own training scripts and, in some cases, separate dependency files.
• Large datasets, caches, and generated training outputs are intentionally kept out of version control.