numta 0.2.0

NumPy-based Technical Analysis library with focus on performance

numta

Pure Python technical analysis library. A modern, high-performance alternative to TA-Lib with zero C dependencies.

📖 Documentation | 🚀 Quick Start | 📚 API Reference

Highlights

• Pure Python: No C compiler required, works everywhere Python runs
• Fast: 5-10x speedup with optional Numba JIT compilation
• GPU Batch Processing: Run any indicator on 26K+ tickers simultaneously with CUDA
• Complete: 130+ indicators, 60+ candlestick patterns, chart pattern detection
• Modern: Pandas integration, streaming support, Jupyter visualization

Installation

# Basic installation
pip install numta

# With Numba for 5-10x speedup
pip install "numta[numba]"

# With GPU batch processing (NVIDIA CUDA)
pip install "numta[gpu]"

# With pandas integration
pip install "numta[pandas]"

# Full installation with all features
pip install "numta[full]"

# From source
git clone https://github.com/deepentropy/numta.git
cd numta
pip install -e .

Quick Start

import numpy as np
from numta import SMA, EMA, RSI

# Create sample price data
close_prices = np.random.uniform(100, 200, 100)

# Calculate indicators
sma = SMA(close_prices, timeperiod=20)
ema = EMA(close_prices, timeperiod=12)
rsi = RSI(close_prices, timeperiod=14)

Features

Technical Indicators

numta provides 130+ technical indicators across multiple categories:

• Overlap Studies: SMA, EMA, DEMA, TEMA, WMA, BBANDS, KAMA, MAMA, T3, SAR
• Momentum: RSI, MACD, STOCH, ADX, CCI, MFI, ROC, MOM, WILLR
• Volume: OBV, AD, ADOSC
• Volatility: ATR, NATR, TRANGE
• Cycle: Hilbert Transform functions
• Statistical: LINEARREG, STDDEV, VAR, CORREL, BETA

See FUNCTION_IMPLEMENTATIONS.md for the complete list.

Pandas Integration

import pandas as pd
import numta  # Auto-registers the .ta accessor

df = pd.DataFrame({
    'open': [...], 'high': [...], 'low': [...],
    'close': [...], 'volume': [...]
})

# Calculate and return as Series
sma = df.ta.sma(timeperiod=20)

# Append indicators to DataFrame
df.ta.sma(timeperiod=20, append=True)   # Adds 'SMA_20'
df.ta.rsi(timeperiod=14, append=True)   # Adds 'RSI_14'
df.ta.macd(append=True)                 # Adds MACD columns

Pattern Recognition

Candlestick Patterns

from numta import CDLDOJI, CDLENGULFING, CDLHAMMER

# Returns +100 (bullish), -100 (bearish), or 0 (no pattern)
doji = CDLDOJI(open_, high, low, close)
engulfing = CDLENGULFING(open_, high, low, close)

# Via pandas accessor
df.ta.cdldoji(append=True)
df.ta.cdlengulfing(append=True)

Chart Patterns

from numta import (
    detect_head_shoulders, detect_double_top,
    detect_triangle, detect_wedge, detect_flag
)

# Detect patterns with confidence scores
patterns = detect_head_shoulders(highs, lows, order=5)

# Via pandas accessor
patterns = df.ta.find_patterns(pattern_type='all')
harmonics = df.ta.find_harmonic_patterns()

GPU Batch Processing

Process thousands of tickers simultaneously on NVIDIA GPUs:

import numpy as np
from numta import SMA_batch, RSI_batch, CDLDOJI_batch

# 2D arrays: (num_tickers, num_bars)
close = np.random.uniform(50, 150, (10000, 500))
high = close + np.random.uniform(0, 5, (10000, 500))
low = close - np.random.uniform(0, 5, (10000, 500))
open_ = close + np.random.uniform(-2, 2, (10000, 500))

# Run on all 10,000 tickers at once — one CUDA thread per ticker
sma = SMA_batch(close, timeperiod=20)        # shape: (10000, 500)
rsi = RSI_batch(close, timeperiod=14)        # shape: (10000, 500)
doji = CDLDOJI_batch(open_, high, low, close) # shape: (10000, 500)

All 128 batch functions (*_batch) mirror the CPU API exactly — same parameters, same output values. Requires an NVIDIA GPU with CUDA support.

Streaming/Real-Time

from numta.streaming import StreamingSMA, StreamingRSI, StreamingMACD

# Create streaming indicators
sma = StreamingSMA(timeperiod=20)
rsi = StreamingRSI(timeperiod=14)

# Process streaming data
for price in price_stream:
    sma_value = sma.update(price)
    rsi_value = rsi.update(price)
    
    if sma.ready and rsi.ready:
        print(f"SMA: {sma_value:.2f}, RSI: {rsi_value:.2f}")

Visualization

# Install visualization support
# pip install "numta[viz]"

from numta.viz import plot_ohlc, plot_with_indicators

# Basic candlestick chart
chart = plot_ohlc(df, volume=True)

# Chart with indicators
chart = plot_with_indicators(df, indicators={'SMA_20': sma_data})

# Plot detected patterns
chart = df.ta.plot(patterns=patterns)

Documentation

📖 Full Documentation - Comprehensive API reference and guides.

Notebooks

• Getting Started
• Technical Indicators Guide
• Candlestick Patterns
• Chart Patterns
• Harmonic Patterns
• Streaming Indicators
• Visualization
• Performance Optimization

Performance

numta uses optimized algorithms and optional Numba JIT compilation:

Implementation	Speed vs Default	Requirements
numpy (default)	1.0x (baseline)	None
cumsum	~3x faster	None
numba	5-10x faster	pip install numba
GPU batch	100x+ faster (multi-ticker)	NVIDIA GPU + pip install "numta[gpu]"

from numta import SMA_auto, SMA_cumsum

# Automatic backend selection
sma = SMA_auto(close_prices, timeperiod=30, backend='auto')

# Or choose specific backend
sma_fast = SMA_cumsum(close_prices, timeperiod=30)

For batch processing across thousands of tickers, the GPU backend provides massive parallelism — each CUDA thread processes one ticker's full time series. See GPU Batch Processing for details.

API Reference

See FUNCTION_IMPLEMENTATIONS.md for detailed implementation status of all indicators.

Requirements

• Python >= 3.8
• NumPy >= 1.20.0

Optional Dependencies

Feature	Package	Installation
Performance	numba >= 0.56.0	pip install "numta[numba]"
GPU Batch	numba-cuda >= 0.1.0	pip install "numta[gpu]"
Pandas	pandas >= 1.3.0	pip install "numta[pandas]"
Visualization	lwcharts >= 0.1.0	pip install "numta[viz]"
All features	-	pip install "numta[full]"

Project Structure

numta/
├── src/numta/
│   ├── __init__.py
│   ├── api/                 # Indicator implementations
│   │   └── batch.py         # GPU batch API (128 *_batch functions)
│   ├── cpu/                 # Numba-optimized CPU kernels
│   ├── gpu/                 # CUDA GPU kernels for batch processing
│   ├── patterns/            # Chart pattern detection
│   ├── streaming/           # Real-time indicators
│   ├── viz/                 # Visualization (lwcharts)
│   ├── pandas_ext.py        # DataFrame accessor
│   ├── backend.py           # Backend selection
│   ├── benchmark.py         # Performance tools
│   └── optimized.py         # Optimized implementations
├── notebooks/               # Example notebooks
├── tests/                   # Test suite
├── pyproject.toml
└── README.md

Testing

# Install test dependencies
pip install -e ".[dev]"

# Run tests
pytest

# Run with verbose output
pytest -v

Contributing

Contributions are welcome! Please:

1. Fork the repository
2. Create a feature branch
3. Implement your changes with tests
4. Ensure all tests pass
5. Submit a pull request

License

MIT License - see LICENSE file for details.

Acknowledgments

This project implements technical analysis algorithms based on publicly available mathematical formulas. We acknowledge:

• TA-Lib - The original Technical Analysis Library
  • Website: https://ta-lib.org/
  • Python wrapper: https://github.com/TA-Lib/ta-lib-python
  • License: BSD 3-Clause

numta is an independent implementation and is not derived from TA-Lib's source code. All code is original work licensed under the MIT License.

Support

For issues, questions, or contributions: https://github.com/deepentropy/numta/issues

Citation

@software{numta,
  title={numta: NumPy-based Technical Analysis Library},
  author={numta contributors},
  url={https://github.com/deepentropy/numta},
  year={2025}
}