adelic-signalforge 0.1.0

P-adic multiscale signal processing pipeline

SignalForge

Multiscale signal processing pipeline built on the p-adic divisibility lattice. Transforms raw sequential data into structurally invariant feature tensors.

---

Install

git clone https://github.com/adelic-ai/signalforge
cd signalforge
uv sync

---

Quick start

uv run signalforge demo                        # EEG seizure detection demo
uv run signalforge neighborhood 36 6           # p-adic arithmetic viewing box
uv run signalforge plan eeg                    # sampling plan for EEG
uv run signalforge run eeg path/to/data.csv    # run on your data

---

Bring your own data

SignalForge ships with two domains: intermagnet (geomagnetic observatory) and eeg (clinical EEG). To add your own, write a single Python file in signalforge/domains/.

• docs/overview.md — what it is, the mathematics, the name
• docs/domain_guide.md — how to write a domain
• docs/data_guide.md — downloading data, running examples, AI-assisted preprocessing
• CONTRIBUTING.md — submitting a domain or pipeline change

The short version: declare a horizon and a grain, let the lattice enumerate valid windows, select the ones meaningful for your data, return a SamplingPlan. Everything downstream runs without modification.

---

Pipeline

Stage	Name	Input → Output
0	Ingest	Raw source data → CanonicalRecord
1	Plan	Declarations → SamplingPlan
2	Materialize	CanonicalRecords → BinnedRecord
3	Measure	BinnedRecords → Surface (time × scale grid)
4	Engineer	Surface → FeatureTensor
5	Assemble	FeatureTensors → FeatureBundle (ML-ready)

Domain knowledge lives exclusively in signalforge/domains/. Stages 1–5 are identical across all data sources.

---

Result: EEG seizure detection

Dataset: CHB-MIT Scalp EEG, patient chb01, recording chb01_03. Input: RMS across 23 channels, 1-second bins. No labels, no seizure times, no EEG-specific processing.

Peak z-score: 13.96σ at the 30-second scale, within the annotated seizure window (seconds 2996–3036).

Full result: docs/discovery_eeg_seizure_detection.md

---

License

Business Source License 1.1. See LICENSE.

Free for non-commercial use (research, personal projects, evaluation). Commercial use requires a license from Adelic — contact [e-mail].

Converts to Apache 2.0 on 2029-03-22.

---

Citation

If you use the CHB-MIT EEG data, please cite:

Shoeb AH, Guttag JV. Application of machine learning to epileptic seizure detection. In: Proceedings of the 27th International Conference on Machine Learning (ICML). 2010.

Raw data: CHB-MIT Scalp EEG Database v1.0.0 https://physionet.org/content/chbmit/1.0.0/