EEG Course Project

This repository contains runnable code and final prediction files for the shared EEG course project across five datasets:

• BCIC2A
• CHINESE
• MDD
• SEED
• SLEEP

Dataset files and generated intermediate artifacts are not committed because they are large. The code expects the course data root to be available at:

/mnt/dataset3/panxy/course/project1_data/course project/course project

On the Lab 212 host, run the project from:

/mnt/dataset4/yichen/ML_practice&homework/course_project_1

The recommended Python environment on the Lab 212 host is timesfm2.

Final Submission Files

The final prediction files required by the course are tracked in submission/:

submission/BCIC2A.txt
submission/CHINESE.txt
submission/MDD.txt
submission/SEED.txt
submission/SLEEP.txt

Each file contains one integer class label per line, without header or filename columns. The row order is the fixed test DataLoader order with shuffle=False.

Expected row counts:

Dataset	Rows	Classes
BCIC2A	360	0-3
CHINESE	200	0-1
MDD	800	0-1
SEED	450	0-2
SLEEP	1945	0-4

Main Reproduction Commands

Run the five-dataset SOTA-first pipeline on the Lab 212 host:

bash run_sota_five_pipeline.sh

The script runs the per-dataset pipelines, validates candidate predictions, and updates outputs/submission/ when a candidate beats the current valid baseline.

Individual scripts:

python run_sota_bcic2a.py
python run_sota_seed.py
python run_sota_chinese.py
python run_sota_mdd.py
python run_sota_sleep.py
python run_sota_finalize.py --update-submission

Current Final Selection

Dataset	Validation balanced accuracy	Final method
BCIC2A	0.6028	FBCSP wide overlap CSP + ExtraTrees + MIBIF
SEED	0.4556	beta/gamma graph-inspired features + Ridge
CHINESE	0.7100	covariance and band statistics + RandomForest
MDD	0.9328	retained strongest CBraMod baseline
SLEEP	0.7594	band/covariance statistics + ExtraTrees

Repository Layout

• run_sota_*.py: final per-dataset pipelines.
• run_sota_five_pipeline.sh: one-command five-dataset run.
• run_sota_finalize.py: conservative final-selection and submission updater.
• sota_common.py: shared data loading, feature extraction, validation, and output utilities.
• run_all_cbramod.py, run_cbramod_pooling_ablation.py, run_cbramod_author_downstream.py: CBraMod baseline and ablation pipelines.
• docs/: method notes, adaptation notes, and model inventory.
• submission/: final five .txt files for course submission.

Files Intentionally Not Tracked

The following are generated or too large for GitHub and are ignored:

• artifacts/
• outputs/
• .codex_runs/
• model weights and data files such as *.h5, *.pt, *.pth, *.npz