Comparison with Other Packages

Quick comparison of fastcpd-python with other change point detection libraries.

Python Packages

ruptures - Comprehensive offline change point detection

• Pure Python (no C++ compilation)

• Multiple algorithms: PELT, Binary Segmentation, Bottom-Up, Window-based

• Mature codebase with large user base

• Extensive documentation

fastcpd-python - This package

• GLM support (Binomial, Poisson)

• Time series models (AR, ARMA, VAR, GARCH)

• SeGD algorithm for large datasets

• Requires C++ (Armadillo)

R Packages

changepoint - Established R package

• Mature, well-tested

• CRAN distribution

• Classical methods (PELT, BinSeg)

fastcpd - Original R implementation

• PELT + SeGD algorithms

• CRAN distribution

• Native R integration

Feature Comparison

Feature	fastcpd-py	ruptures	R fastcpd	changepoint
PELT	Yes	Yes	Yes	Yes
Binary Segmentation	No	Yes	No	Yes
SeGD	Yes	No	Yes	No
Mean/Variance	Yes	Yes	Yes	Yes
GLM	Yes	No	Yes	No
Time Series	Yes	No	Yes	No
Nonparametric	Yes	Yes	Yes	No
Installation	pip + C++	pip only	CRAN	CRAN

When to Use Each

Use fastcpd-python for:

• GLM or time series models in Python

• Large datasets (SeGD algorithm)

• Comprehensive evaluation metrics

Use ruptures for:

• Pure Python without C++ dependencies

• Binary segmentation or bottom-up algorithms

• Established Python package

Use R fastcpd for:

• R ecosystem integration

• CRAN distribution preference

Use changepoint for:

• Mature R package

• Classical methods in R

Migration Examples

From ruptures

# ruptures
import ruptures as rpt
algo = rpt.Pelt(model="rbf").fit(signal)
result = algo.predict(pen=10)

# fastcpd-python
from fastcpd.segmentation import rbf
result = rbf(signal, beta=10.0)
change_points = result.cp_set

From R fastcpd

# R
library(fastcpd)
result <- fastcpd.mean(data, beta="MBIC")

# Python
from fastcpd.segmentation import mean
result = mean(data, beta="MBIC")

Summary

Choose based on:

1. Programming language (Python vs R)

2. Model requirements (standard vs GLM vs time series)

3. Installation constraints (pure Python vs C++)

4. Algorithm needs (PELT, SeGD, Binary Segmentation)