zeon-ta

Technical analysis for Python — the widely used classics (RSI, MACD, Bollinger Bands, Ichimoku, and the rest of the standard toolkit) alongside modern, academically-sourced tools most TA libraries skip: Ehlers' cycle-analysis filters, the Hurst exponent, wavelet-based denoising and multi-scale volatility, OHLC volatility estimators standard in quantitative finance (Parkinson, Garman-Klass, Rogers-Satchell, Yang-Zhang), and a causal cross-wavelet lead-lag transform.

Formulas follow standard, widely published technical-analysis definitions where one exists. Where a formula's own academic paper is the source instead, or where a candidate indicator turned out to have no single agreed-on formula across implementations, the docstring says which and why.

Why another TA library

• Broad on purpose. 145 indicators across 8 categories (moving averages, oscillators, volatility, trend, volume, statistics) — not just the popular dozen.
• Classic and modern, both formula-verified. Every indicator — whether it is RSI or a MODWT wavelet-variance decomposition — cites what its formula was checked against, and a proposed indicator with no single agreed-on formula across sources is declined outright rather than guessed at (documented in CHANGELOG.md either way).
• No build step. Every dependency ships prebuilt wheels, so pip install just works — everywhere, including on ARM Macs and in slim containers.
• One contract, every indicator. Pass a Series, an array or a list; get pandas back with your index intact and the same length as your input. Warm-up bars are NaN, never trimmed, so nothing silently shifts under a backtest.
• Two ways to call it. A functional API and a .zta DataFrame accessor that routes to the exact same code — verified equal by tests, not by convention.
• Documented honestly. Every indicator's page states its pitfalls, including where an output contains look-ahead information and what to do about it.
• Measured, not assumed, performance. Every indicator is benchmarked at up to 1M bars, with real numbers and methodology in BENCHMARKS.md — most complete in low milliseconds even at that size.

Install

pip install zeon-ta

Or install the latest commit straight from GitHub, ahead of the next release:

pip install git+https://github.com/selimozbas/zeon-ta.git

Or clone and install locally:

git clone https://github.com/selimozbas/zeon-ta.git
cd zeon-ta
pip install .

Requires Python 3.12+.

Quick start

import pandas as pd
import zeonta

df = pd.read_csv('ohlcv.csv', parse_dates=['date']).set_index('date')

# Functional
rsi = zeonta.rsi(df['close'], length=14)
bands = zeonta.bbands(df['close'], length=20, std=2)

# Accessor — identical results
rsi = df.zta.rsi(length=14)
trend = df.zta.supertrend(length=10, multiplier=3)

# Discover everything that is available
print(zeonta.list_indicators())

More in examples/, runnable directly against a committed sample dataset.

Output contract

Single-line indicators return a named Series; multi-line ones return a DataFrame whose column names carry the settings used (RSI_14, MACD_12_26_9, SUPERT_10_3.0). ichimoku additionally returns the part of the cloud that projects past the last bar, rather than discarding it.

Documentation

The full indicator reference — 145 indicators across 8 categories, each with its formula, parameters, worked examples and (where one exists) the external source it was verified against — is published at:

https://selimozbas.github.io/zeon-ta/

It's generated straight from the code and from actually running every example (see tools/gen_docs.py), so it never drifts out of sync with what's installed. Browse it locally under docs/ instead if you'd rather not leave the repo.

zeonta.cross_asset.wavelet_lead_lag(close_a, close_b, period=20) compares two independent price series — which one is leading the other, and by how much, at a chosen timescale — via a causal Morlet Cross-Wavelet Transform (Torrence & Compo, 1998). It isn't in list_indicators() or the .zta accessor: every registered indicator assumes one asset's own OHLCV columns, and a second, independent series doesn't fit that contract. Import and call it directly; see its own docstring for the full method and a documented lag-estimate caveat.

Roadmap

No fixed release calendar — see ROADMAP.md for the current direction (more indicator sources, an optional acceleration path, backtest-engine integration examples) and what's deliberately out of scope.

Development

pip install -e ".[dev]"
pytest                      # test suite
ruff check . && mypy src/   # lint and types
python tools/gen_docs.py    # regenerate the docs

Documentation is generated: prose lives in tools/docs_content.py, while parameter tables, column names and example output are taken from the code itself and from actually running each example. A test fails if the committed files drift.

See CONTRIBUTING.md for the full workflow, and docs/methodology.md for how a formula gets verified before it's implemented. This project follows a Code of Conduct; see SECURITY.md to report a vulnerability privately.

License

MIT — see LICENSE.