All work
Open source2025Maintainer
Sensor data cleaning that shows its work
A small Python library for cleaning noisy time-series from lab equipment, where every transformation is logged and reversible.
Signal
The problem
Preprocessing is where quiet mistakes live. A dropped outlier or a resampled timestamp changes a conclusion, and by the time the plot exists nobody remembers which filters ran in what order.
The approach
Each operation returns both the transformed data and a record of what it did. Stack the operations and you get a pipeline that can explain itself and, where the maths allows, walk itself backwards.