tisa-finance 0.2.2

Transform-Invariant Segment Alignment (TISA) for financial time series

TISA

Transform-Invariant Segment Alignment for financial time series

TISA is a novel algorithm for aligning and comparing financial time series in a way that is invariant to non-linear transformations (like different volatility regimes or trends) and robust to noise. It combines segmentation with dynamic programming to find the optimal alignment between two series.

Theory

Standard measures like Euclidean distance or Dynamic Time Warping (DTW) often fail on financial data because they are sensitive to amplitude scaling and noise. TISA addresses this by:

1. Segmentation: Breaking the time series into linear segments to reduce noise and dimensionality.
2. Transform Invariance: Allowing each segment in the source series to be transformed (scaled and shifted) to match the target series.
3. Dynamic Programming: Finding the globally optimal sequence of segments and transforms that minimizes the alignment cost.

For more details, see the documentation.

Install

Requires Python 3.9+.

# Create a virtual environment
python3 -m venv .venv
source .venv/bin/activate

# Install TISA
pip install -U pip
pip install -e .[dev]

Optional: create a .env file with your Polygon.io API key for data fetching:

POLYGON_API_KEY=YOUR_KEY_HERE

Quick Start

Python API

import numpy as np
from tisa.distance import TISADistance

# Generate synthetic data
x = np.cumsum(np.random.randn(252))
y = x * 1.5 + 10  # Scaled and shifted version

# Compute distance
dist = TISADistance().pairwise(x, y)
print(f"TISA Distance: {dist:.4f}")

# Get detailed alignment info
detail = TISADistance().detailed(x, y)
print(f"Best Transform: {detail['best_transform']}")

CLI

Fetch data (auto-uses Polygon if POLYGON_API_KEY is present, else yfinance):

tisa fetch --tickers SPY,QQQ --interval 1d --start 2022-01-01 --end 2025-01-01 --source auto

Compute distance / alignment from CSV files:

tisa distance --fileA data/SPY_1d.csv --fileB data/QQQ_1d.csv
tisa align --fileA data/SPY_1d.csv --fileB data/QQQ_1d.csv --out reports/align_SPY_QQQ.json

Benchmarks

TISA comes with a benchmarking suite to evaluate performance against other metrics.

# Run daily benchmark
tisa bench --config tisa/benchmarks/multi_daily.yaml --out reports/benchmark_daily

See tisa/benchmarks/README.md for more info.

Performance Benchmarks

Track TISA performance over time using pytest-benchmark:

# Run performance benchmarks
pytest tisa/tests/test_performance.py --benchmark-only

# Compare with previous runs
pytest tisa/tests/test_performance.py --benchmark-compare

# Save baseline
pytest tisa/tests/test_performance.py --benchmark-save=baseline

Contributing

We welcome contributions! Please see CONTRIBUTING.md for details.

License

MIT