Python package for fast and flexible univariate trend filtering with C++ backend for high performance.

Trend filtering is a method for nonparametric regression that fits a piecewise polynomial function to data. This package provides efficient implementations of trend filtering algorithms with the following features:

• Fast C++ backend: High-performance implementations using modern C++
• Python interface: Easy-to-use Python API built with pybind11
• Flexible algorithms: Support for different orders of trend filtering
• Cross-validation: Built-in cross-validation for parameter selection
• Efficient solvers: Optimized linear system solvers for large datasets

pip install git+https://github.com/jiapivialiu/trendfilter-py.git

git clone https://github.com/jiapivialiu/trendfilter-py.git
cd trendfilter-py
pip install .

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
from trendfilter import TrendFilter, CVTrendFilter

# Generate sample data
n = 100
x = np.linspace(0, 1, n)
true_signal = np.sin(2 * np.pi * x) + 0.5 * np.cos(4 * np.pi * x)
y = true_signal + 0.2 * np.random.randn(n)

# Fit trend filter
tf = TrendFilter(order=2, lambda_reg=0.1)
tf.fit(y)
y_fit = tf.predict()

# Cross-validation for parameter selection
cv_tf = CVTrendFilter(order=2)
cv_tf.fit(y)
y_fit_cv = cv_tf.predict()

Check out the README_plot script or the basic_usage for visualizing the output. Preview of the output as below:

• Trend Filtering: Fast implementation of univariate trend filtering
• Cross-Validation: Automated parameter selection via cross-validation
• Multiple Orders: Support for different polynomial orders (0, 1, 2, ...)

• C++ Backend: High-performance C++ implementations
• Efficient Solvers: Specialized linear system solvers
• Memory Optimized: Efficient memory usage for large datasets

• Python >= 3.7
• NumPy >= 1.18.0
• SciPy >= 1.5.0