Add 3W Dataset overview: sensor availability and cleaning impact

[GabrielRozo123]

What this adds

A new dataset overview notebook at dataset/demos/GabrielRozo/main.ipynb that quantifies how often each process sensor is actually available across the 3W Dataset, broken down by event class.

Key findings

• Several sensors (P-JUS-BS, P-MON-CKGL, P-MON-SDV-P, PT-P, QBS) are effectively never recorded in the sampled events.
• A small, reliable backbone exists (P-PDG, P-TPT, P-MON-CKP), present in most events of most classes.
• Sensor availability is class-dependent, for example, P-JUS-CKP and T-MON-CKP appear almost only in classes 5 and 6 which suggests detection models cannot assume a fixed feature set across events.

Method & reproducibility

Availability is measured as the fraction of non-missing readings per sensor, averaged over a fixed random sample of events per class (fixed seed). During the analysis, the toolkit's CleanSignals step produced run-dependent column selections for the same event, so the overview relies on the raw, deterministic missing-data fraction instead.

CLA

I have read and agree to the Contributor License Agreement (ICLA) and the Code of Conduct.

[ricardoevvargas]

Hi, @GabrielRozo123.

Thank you for submitting this overview, which will be reviewed as soon as possible.