wavelet feature extraction

Wavelet feature extraction for ML pipelines.

FerroWave turns wavelet decompositions into features that can move from research notebooks to Rust inference and analytics systems.

Problem

Why this needs production signal processing

Time-domain features miss scale-localized behavior that matters in noisy signals.

Notebook feature code often breaks when moved into lower-latency production systems.

ML pipelines need repeatable feature generation across research and deployment environments.

Workflow

Classical or adaptive transforms expose behavior across time, scale, frequency, and mode.

Feature vectors, scaleograms, coefficients, energies, ridges, and streaming summaries can feed downstream models.

The same transform path can support batch research and production feature extraction.

Fit

✓

The public site explicitly highlights AI and ML feature extraction.

✓

FerroWave supports multi-scale vectors, scaleogram tensors, and streaming features.

✓

Reference-pinned behavior helps preserve research-to-runtime equivalence.

Questions

No. Financial signals are one use case; the feature-extraction story applies to broader time-series ML pipelines.

Yes. Streaming transforms and online inference features are part of the public positioning.

Time-series feature extraction in Rust using wavelets, adaptive decompositions, scale-localized summaries, and streaming signal features.

Signal preprocessing for ML with wavelet denoising, multi-scale decomposition, adaptive modes, scaleograms, and Rust deployment paths.

Streaming wavelet transform workflows for low-latency tick, sensor, and time-series systems that need online multi-scale features.