jbrandonp/havok-toolbox — explained in plain English
Analysis updated 2026-05-18
Analyze a time series file through a drag-and-drop web dashboard without writing any code.
Detect early warning signals before a system undergoes a sudden regime shift, such as in EEG, financial, or sensor data.
Plug a regime-shift detector into an existing scikit-learn machine learning pipeline.
Monitor a live data stream and trigger alerts when risk of a sudden shift rises.
| jbrandonp/havok-toolbox | 0xallam/my-recipe | 0xhassaan/nn-from-scratch | |
|---|---|---|---|
| Stars | 0 | — | 0 |
| Language | Python | Python | Python |
| Last pushed | — | 2022-11-22 | — |
| Maintenance | — | Dormant | — |
| Setup difficulty | moderate | moderate | moderate |
| Complexity | 4/5 | 2/5 | 4/5 |
| Audience | researcher | general | developer |
Figures from each repo's GitHub metadata at analysis time.
Base install is a simple pip install, but GPU, streaming, and transformer features each require separate optional extras.
havok-toolbox is a Python library that tries to spot the warning signs of a sudden shift in a system before that shift actually happens. It is built around an algorithm called HAVOK, published by Brunton, Brunton, Proctor and Kutz in a 2017 Nature Communications paper. The idea is that many real-world signals, such as an EEG reading before a seizure, a stock price before a crash, a climate measurement before a tipping point, or sensor data before a machine breaks down, contain a hidden pattern that behaves like small unpredictable nudges to an otherwise steady system. HAVOK pulls that hidden pattern, called the forcing signal, out of a plain time series, and it tends to spike before the raw data shows any visible change. The project offers three ways in. For people who do not want to write code, there is a Streamlit dashboard you can open in a browser, where you drag and drop a CSV file and get an analysis and downloadable report. For quick checks from a terminal, there is a command line tool that analyzes a file and writes results to a new CSV. For developers, there is a full Python API with an estimator that follows the same fit and transform pattern used by scikit-learn, so it can slot into existing machine learning pipelines and tools like GridSearchCV. Beyond the basic algorithm, the toolbox adds automatic parameter tuning, support for analyzing multiple signal channels at once such as multi-sensor or multi-asset data, automated hyperparameter search using Optuna, and optional hybrid models that combine HAVOK with a transformer neural network or an echo state network for forecasting. It also includes uncertainty estimation methods, tools for measuring how chaotic a system is, a streaming mode for live data feeds with alerting, and an optional privacy-preserving mode for combining data across multiple institutions without sharing raw data directly. GPU acceleration and a faster CSV loader are available as optional extras. The project is released under the MIT license and includes an automated test suite.
A Python toolbox that detects early warning signs of sudden shifts in time series data, such as before a market crash, a seizure, or an equipment failure.
Mainly Python. The stack also includes Python, scikit-learn, PyTorch.
Use freely for any purpose, including commercial use, as long as you keep the copyright notice.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly researcher.
This repo across BitVibe Labs
double-check against the repo, no cap.