Overview
Where the MCP server and adaptive-interface work are about live brain state, this is the patient, after-the-fact half: a kit of Python analysis tools that loads sessions from a SQL Server database or CSV and hunts for structure -; which band combinations cluster together, which correlate with which activity, and how patterns evolve over the length of a session.
The motivating dataset was wonderfully mundane: brainwaves recorded while playing Monopoly, used to ask whether you could spot "strategic thinking" states in the signal.
Background
Once there were stacks of recorded CSVs (and a SQL Server database, mind, with
sensor and session tables), the natural question was whether different activities leave
distinct brainwave "fingerprints". That turned into a fairly serious analysis stack -;
and, because the data got big, eventually a GPU-accelerated hierarchical analyzer to keep
the heavier algorithms tractable.
How It Works
At the core is a pattern analyzer that works across the eight bands (Delta, Theta, Alpha1/2, Beta1/2, Gamma1/2) using a familiar scikit-learn toolbox:
- Clustering -; KMeans and DBSCAN to find natural groupings of brain states without labels; PCA and t-SNE to make the high-dimensional band space visible.
- Correlation & classification -; correlation analysis between bands and activities, plus Random Forest classifiers (with cross-validation) to test whether an activity is predictable from the signal.
- Temporal patterns -; peak-finding and time-series analysis to catch how states shift across a session rather than treating each sample as independent.
- Storage layer -; loads equally from CSV or a SQL Server connection, and writes discovered patterns and activity fingerprints back out.
A separate, more exotic module reaches for the heavier neuroscience methods -; phase-amplitude coupling, entropy measures, microstate and functional-connectivity analysis -; to look for cognitive-state signatures the basic stats would miss.
Where It Landed
Archived as a working analysis suite. It ran end to end -; there are exported activity fingerprints and analysis results in the repo -; with a master pipeline that processes all sessions and emits an HTML report. The honest caveat is the data: single-sensor consumer EEG with lots of artifacts, so "fingerprints" are suggestive, not conclusive.
- Clustering, correlation, classification, and temporal analysis over recorded sessions.
- CSV and SQL Server ingestion; exported fingerprints and a master HTML report.
- Optional GPU-accelerated path for the larger datasets and heavier algorithms.