EEGNetReplication data-science pipeline

Replication of results of the original EEGNet paper. We are focused on the SMR test replication specifically

In this project following steps were done for the replication process:

• fetches data: BCI Competition IV; Dataset 2a,
• pre-process fetched data,
• trains a classifier (CNN Model) to predict SMR actions,
• logs everything to app.log.

Quickstart

python -m venv .eegnetenv
# Windows: .eegnetenv\Scripts\activate
# macOS/Linux: source .eegnetenv/bin/activate

pip install -e ".[ds,test,lint]"

1. Fetch data (cached into data/raw/) from kaggle

python -m eegnet_repl.fetch --src kaggle

Alternative:

python -m eegnet_repl.fetch --src moabb

2. Build dataset + train model

python -m eegnet_repl.train --test-size 0.2 --seed 42

3. Run UI

python -m eegnet_repl.ui

What students should implement (good Git issues)

• Add 2 more engineered features (e.g., log-transform, earth-only close approaches)
• Add a second model (e.g., RandomForest) and compare results
• Add a saved confusion-matrix figure under reports/
• Add one more test (e.g., “dataset has no negative diameters”, “model predicts probabilities in [0,1]”)

Notes

• DEMO_KEY is fine for small experiments but has low rate limits; students can generate their own API keys on NASA Open APIs.