Installation

Requirements

Pandas TA Classic requires:

Python (the latest stable version plus the prior 4 minor versions)
pandas
numpy

Note

Python version support follows a rolling policy: the latest stable Python version plus 4 preceding minor versions, currently Python 3.11–3.15. The window is set by LATEST_PYTHON_VERSION in the CI workflow, which a maintainer raises after a new Python release.

Optional Dependencies

For enhanced functionality, consider installing:

TA-Lib: Optional — all 62 CDL patterns work natively without it (see Installing TA-Lib below)
tulipy: Optional — only needed to regenerate the frozen oracle golden file (tests/fixtures/generate_tulipy_oracle.py on CPython <3.12); test_oracle_tulipy.py compares against the committed snapshot and no longer imports tulipy at test time
yfinance / alpha-vantage: pandas-ta-classic does not download data. Install a data provider yourself (pip install yfinance) and pass its OHLCV DataFrame in (see examples/fetch_market_data.py)
vectorbt / backtrader / backtesting: For backtesting integration (pip install pandas-ta-classic[backtest])
pytest + Hypothesis: For running the test suite and property-based tests (pip install pandas-ta-classic[test])

Installation Methods

Pandas TA Classic supports both modern uv and traditional pip package managers.

Stable Release (Recommended)

Install the latest stable release from PyPI:

Using uv (recommended - faster):

uv pip install pandas-ta-classic

Using pip:

pip install pandas-ta-classic

Latest Development Version

Install the most recent version with all latest features and bug fixes:

Using uv:

uv pip install git+https://github.com/xgboosted/pandas-ta-classic

Using pip:

pip install -U git+https://github.com/xgboosted/pandas-ta-classic

Development Installation

For contributing to the project or testing unreleased features:

Using uv:

# Clone the repository
git clone https://github.com/xgboosted/pandas-ta-classic.git
cd pandas-ta-classic

# Install in editable mode with all dependencies
# (excludes backtest — install it explicitly, see below)
uv pip install -e ".[all]"

# Or install specific dependency groups:
uv pip install -e ".[dev]"      # Development tools
uv pip install -e ".[test]"     # Testing: pytest, Hypothesis
uv pip install -e ".[optional]" # Optional runtime features (tqdm progress bars)
uv pip install -e ".[oracle]"   # Oracle parity lib: TA-Lib (tulipy values are frozen)
uv pip install -e ".[backtest]" # Backtesting: backtesting, vectorbt, backtrader

Using pip:

# Clone the repository
git clone https://github.com/xgboosted/pandas-ta-classic.git
cd pandas-ta-classic

# Install in editable mode with all dependencies
# (excludes backtest — install it explicitly, see below)
pip install -e ".[all]"

# Or install specific dependency groups:
pip install -e ".[dev]"      # Development tools
pip install -e ".[test]"     # Testing: pytest, Hypothesis
pip install -e ".[optional]" # Optional runtime features (tqdm progress bars)
pip install -e ".[oracle]"   # Oracle parity lib: TA-Lib (tulipy values are frozen)
pip install -e ".[backtest]" # Backtesting: backtesting, vectorbt, backtrader

Note

Development Installation Requirements:

Full git repository with history and tags (not a shallow clone)
setuptools-scm is automatically installed as a build dependency
Git tags determine the package version (e.g., 0.3.36.dev1 for development, 0.3.35 for releases)
A GitHub source archive (pip install https://github.com/xgboosted/pandas-ta-classic/archive/main.zip or Download ZIP) carries its version in .git_archival.txt, which git archive fills in
A source copy with neither .git nor that file (for example a Docker COPY without .git) fails to build with setuptools-scm’s LookupError instead of installing as 0.0.0; include .git or set SETUPTOOLS_SCM_PRETEND_VERSION
See the Version Management section in CONTRIBUTING.md for details

Oracle Libraries (TA-Lib and tulipy)

Both TA-Lib and tulipy are fully optional. They serve different roles:

Library	Role	Effect when installed
TA-Lib	Acceleration backend + oracle	Core indicators — native by default, opt-in via `talib=True`; also used by `test_oracle_talib.py` for parity checks
tulipy	Oracle only	Never used as computation backend; only `test_oracle_tulipy.py` uses it to verify native output

Installing TA-Lib

TA-Lib has a dual role: acceleration backend for core indicators (opt-in via talib=True), and parity oracle. The two behavioural areas affected are:

Candlestick patterns (CDL family)

All 62 CDL patterns have native Python implementations that are always used. TA-Lib is never invoked for CDL patterns — the TA-Lib fallback code path in cdl_pattern() is only retained for hypothetical future patterns without a native implementation.

Core indicators (ema, sma, rsi, macd, obv, atr, etc.)

The native implementation is used by default. TA-Lib is opt-in — pass talib=True to any call to use TA-Lib’s implementation instead.

import pandas_ta_classic as ta

# Uses native EMA — default behaviour
ema = df.ta.ema(length=20)

# Use TA-Lib implementation if installed
ema = df.ta.ema(length=20, talib=True)

# CDL patterns — always native, talib= kwarg has no effect here
df = df.ta.cdl_pattern(name="engulfing")       # native
result = df.ta.cdl_pattern(name="hammer")      # native

Installation

Binary wheels (v0.6.5+) bundle the TA-Lib C library — no separate C library setup needed. Versions before 0.6.5 lack binary wheels and require manual C library installation. We require >=0.6.8 so pip install ta-lib works out of the box on Linux, macOS, and Windows (x86_64, arm64):

Using uv:

uv pip install "ta-lib>=0.6.8"

Using pip:

pip install "ta-lib>=0.6.8"

If your platform is not covered by binary wheels, or you prefer a source build, install the TA-Lib C library first. For Conda users:

conda install -c conda-forge libta-lib
conda install -c conda-forge ta-lib

See the TA-Lib Python README for platform-specific instructions and troubleshooting.

Installing tulipy

tulipy is an oracle-only library. It is never used as a computation backend — its sole purpose is test_oracle_tulipy.py, which verifies that native indicator output matches tulipy’s reference values.

Note

Installing tulipy has no effect on indicator behaviour or performance at runtime. It only enables the tulipy oracle test suite.

Using uv:

uv pip install tulipy

Using pip:

pip install tulipy

Or install both oracle libraries at once:

# uv
uv pip install pandas-ta-classic[oracle]

# pip
pip install pandas-ta-classic[oracle]

test_oracle_talib.py is guarded with @unittest.skipUnless and skips automatically when TA-Lib is not installed. test_oracle_tulipy.py compares against tulipy values frozen in tests/fixtures/tulipy_oracle.json, so it runs without tulipy installed.

Installing Optional Dependencies

For complete functionality:

Using uv:

# For stock data download
uv pip install yfinance

# For backtesting
uv pip install backtesting
uv pip install backtrader
uv pip install vectorbt

# For enhanced performance (optional — 6–200× faster recursive indicators on 100,000-bar series)
uv pip install pandas-ta-classic[performance]

# Or install numba directly
uv pip install numba

# Progress bar for strategy() runs (tqdm)
uv pip install pandas-ta-classic[optional]

# The three backtesting libraries above in one go
uv pip install pandas-ta-classic[backtest]

# Every extra (optional, oracle, dev, test, docs, performance)
uv pip install pandas-ta-classic[all]

Using pip:

# For stock data download
pip install yfinance

# For backtesting
pip install backtesting
pip install backtrader
pip install vectorbt

# For enhanced performance (optional — 6–200× faster recursive indicators on 100,000-bar series)
pip install pandas-ta-classic[performance]

# Or install numba directly
pip install numba

# Progress bar for strategy() runs (tqdm)
pip install pandas-ta-classic[optional]

# The three backtesting libraries above in one go
pip install pandas-ta-classic[backtest]

# Every extra (optional, oracle, dev, test, docs, performance)
pip install pandas-ta-classic[all]

Verification

Verify your installation:

import pandas_ta_classic as ta
import pandas as pd

# Create a simple DataFrame
df = pd.DataFrame({'close': [100, 101, 102, 101, 100]})

# Test an indicator
sma = df.ta.sma(length=3)
print(sma)

# List all available indicators
print(f"Available indicators: {len(df.ta.indicators(as_list=True))}")

If this runs without errors, you’re ready to use Pandas TA Classic!