EEG Model Generalization Evaluation

I have a seizure-detection model that already performs well on its original training set. I now need to see how well it generalises to a second, completely separate EEG collection provided in EDF format.

Your task is to plug this new EDF dataset into the existing Python codebase without altering the architecture, hyper-parameters, or training logic. If the raw files require extra steps—channel mapping, resampling, re-referencing, or similar—you may add or tweak preprocessing so the data flows cleanly into the current pipeline, but the model itself must stay untouched.

Once the pipeline runs end-to-end on the new data, compute the ROC-AUC and return:
• the updated preprocessing script or notebook
• a concise report (tables, plots, brief commentary) summarising ROC-AUC and any observations about distributional shift or failure cases

I will supply the repository, a quick start guide, and a sample of the EDF files. Familiarity with MNE-Python, NumPy/Pandas, and scikit-learn will make the job straightforward. If questions come up about electrode naming or file structure, let me know early so we can resolve them quickly.