neerajwagh/eeg-self-supervision

The evaluation directory now includes scripts and model definitions for evaluating self-supervised learning (SSL) models, including a SOTA evaluation script, ablation models (FullSSLNet, ContrastiveTripletNet, BarlowTwinsDBRatioNet, etc.), and dataset utilities. This enables running held-out evaluations for tasks like gender, age, and condition classification on the TUH and Lemon datasets using various SSL architectures.

evaluation · high confidence

The Fine-tune directory now includes a complete supervised learning pipeline for EEG data. This includes model architectures (SOTA\_model, ablation\_models) implementing a Shallow ConvNet and various self-supervised heads (BarlowTwins, Contrastive Triplet), along with a training pipeline (SOTA\_pipeline, ablation\_pipeline) that tracks train/valid metrics (AUC, F1, precision, recall, balanced accuracy) at both sample and patient levels. Supporting modules for dataset handling (datasets.py) and data splitting (utils.py) are also added.

Fine-tune · high confidence

The \supervised\_learning\ directory now includes \linear\_baseline\_train.py\, a script that trains a logistic regression model using scikit-learn's \GridSearchCV\ to find optimal hyperparameters for tasks like gender, age, and condition classification on Tuh and Lemon datasets. It also adds \utils.py\, which provides helper functions for subject-level data splitting, custom cross-validation folding, and patient-level metric aggregation.

_supervised\learning · high confidence

The repository now includes a Dockerfile that sets up a Python 3.8 environment with MNE, PyTorch, and other dependencies, facilitating reproducible execution of the project. Additionally, a new Jupyter notebook (\_00\_generate\_band\_power\_topomap\_image\_from\_timeseries\_signal.ipynb) was added to demonstrate how to extract topomap image arrays from timeseries signals, providing a concrete example for users working with EEG data processing.

(repo-wide) · high confidence

Added new utility modules for EEG signal processing. The \utils/features.py\ module introduces an \eeg\_power\_in\_bands\ function that extracts relative power features across standard brain rhythm bands (delta, theta, alpha, beta, gamma) from MNE Epochs objects, returning data compatible with scikit-learn. The \utils/topomaps.py\ module provides helper functions to convert matplotlib figures to NumPy arrays and apply circular head masks for topographic visualization.

utils · high confidence

Written by watchdog.canine.dev from the codebase's own history, inside the signed delivery this page is composed from.

• CAI 47 → 45 (-2.3)
• Rubric changed (rubric-2026.08.18 → rubric-2026.09.15) — scores are not directly comparable.

• Code Health 94 → 94 (+0.3)
• Architecture 100 → 100 (+0.0)
• Maturity 51 → 44 (-7.1)
• Readiness 17 → 17 (+0.0)
• Security 99 → 93 (-5.8)

• Dependency hygiene not measured — no supported dependency manifest was read
• Duplicated block (11 lines × 2) (Fine-tune/SOTA_model.py)
• Duplicated block (11 lines × 2) (Fine-tune/ablation_models.py)
• Duplicated block (12 lines × 2) (Fine-tune/ablation_models.py)
• Duplicated block (12 lines × 2) (Fine-tune/ablation_models.py)
• Duplicated block (12 lines × 2) (evaluation/linear_baseline_eval.py)
• Duplicated block (12 lines × 3) (Fine-tune/utils.py)
• Duplicated block (13 lines × 2) (Fine-tune/SOTA_pipeline.py)
• Duplicated block (13 lines × 2) (Fine-tune/ablation_models.py)
• Duplicated block (13 lines × 3) (Fine-tune/utils.py)
• Duplicated block (14 lines × 3) (evaluation/SOTA_eval.py)
• Duplicated block (15 lines × 3) (Fine-tune/utils.py)
• Duplicated block (16 lines × 3) (Fine-tune/utils.py)
• Duplicated block (18 lines × 2) (Fine-tune/SOTA_model.py)
• Duplicated block (19 lines × 2) (Fine-tune/ablation_models.py)
• Duplicated block (21 lines × 2) (Fine-tune/SOTA_model.py)
• Duplicated block (24 lines × 2) (Fine-tune/SOTA_model.py)
• Duplicated block (5 lines × 2) (evaluation/linear_baseline_eval.py)
• Duplicated block (6 lines × 2) (Fine-tune/ablation_models.py)
• Duplicated block (7 lines × 2) (evaluation/linear_baseline_eval.py)
• …and 8 more

• Documentation: no usage examples (README.md)
• Duplicated block (10 lines × 2) (Fine-tune/ablation_models.py)
• Duplicated block (10–11 lines × 6) (Fine-tune/utils.py)
• Duplicated block (11 lines × 2) (Fine-tune/ablation_models.py)
• Duplicated block (11 lines × 2) (Fine-tune/ablation_models.py)
• Duplicated block (111 lines × 2) (Fine-tune/SOTA_model.py)
• Duplicated block (12 lines × 2) (Fine-tune/ablation_models.py)
• Duplicated block (13 lines × 2) (Fine-tune/ablation_models.py)
• Duplicated block (13 lines × 3) (Fine-tune/utils.py)
• Duplicated block (13 lines × 3) (evaluation/SOTA_eval.py)
• Duplicated block (13–14 lines × 2) (Fine-tune/SOTA_pipeline.py)
• Duplicated block (14 lines × 2) (Fine-tune/SOTA_model.py)
• Duplicated block (14 lines × 2) (Fine-tune/SOTA_model.py)
• Duplicated block (17 lines × 3) (Fine-tune/utils.py)
• Duplicated block (25 lines × 2) (Fine-tune/SOTA_model.py)
• Duplicated block (30 lines × 2) (Fine-tune/ablation_models.py)
• Duplicated block (31 lines × 2) (Fine-tune/SOTA_model.py)
• Duplicated block (31 lines × 3) (Fine-tune/utils.py)
• Duplicated block (32 lines × 2) (Fine-tune/datasets.py)
• Duplicated block (34 lines × 2) (evaluation/linear_baseline_eval.py)
• …and 39 more

Written by watchdog.canine.dev from the codebase's own history, inside the signed delivery this page is composed from.

neerajwagh/eeg-self-supervision was measured the same way every project in this corpus was: the same rubric, at a pinned commit, with the result published in full. Point a surveyor at a repository you know and see whether you agree with it.