changepy

0.4.0 verified Fri May 01 auth: no python

Changepoint detection on time series in Python. Version 0.4.0 provides algorithms like PELT and binary segmentation for detecting shifts in mean and variance. Low release cadence; last release in 2020.

pip install changepy

Common errors

error ImportError: cannot import name 'pelt' from 'changepy' ↓

cause Outdated or incorrect installation; package may not be installed correctly.

fix

Ensure changepy is installed: pip install --upgrade changepy

error ValueError: The model must be one of 'mean', 'var', or 'meanvar'. ↓

cause Using an invalid model argument in pelt() or binseg().

fix

Use one of the allowed string values: 'mean', 'var', or 'meanvar'.

Warnings

gotcha The PELT implementation expects data to be an array-like of floats. Passing integer arrays may cause type errors. ↓

fix Convert your data to float: arr = np.array(data, dtype=float).

deprecated No active development since 2020; consider using ruptures for maintained changepoint detection. ↓

fix Switch to ruptures (pip install ruptures) for more algorithms and active maintenance.

Imports

pelt
```
from changepy import pelt
```
binseg
```
from changepy import binseg
```

Quickstart

Basic usage of PELT changepoint detection on a synthetic time series.

import numpy as np
from changepy import pelt

# Generate example time series with one changepoint
np.random.seed(42)
ts = np.concatenate([np.random.normal(0, 1, 50),
                     np.random.normal(5, 1, 50)])
# Detect changepoints
changepoints = pelt(ts, penalty=2*np.log(len(ts)))
print(changepoints)  # Expected: [49] (0-indexed), typically