changepoint_cython

A cython version of the changepoint and changepoint.np R package, see https://github.com/cran/changepoint and https://github.com/cran/changepoint.np.

Implemented algorithms:

• Binary Segmentation, PELT, and Segmentation Neigborhood: monovariate and multivariate time series.
• Non parametric PELT: monovariate variate.

Model implemented for the monovariate case :

• Normal (mean, variance, or mean and variance).
• Poisson (mean and variance).
• Exponential (mean and variance).

Model implemented for the multivariate case :

• Normal (mean).

Working on Windows/Linux and python3.x

Install

PyPI

Source available on https://pypi.org/project/changepoint-cython/

pip install changepoint-cython

From source

Download and run:

pip3 install -r requirements.txt
pip3 install .

Tests

Synthetic examples

python3 test_segmentation.py

python3 test_segmentation_multiple.py

Test on well bore data

See notebook "segmentation example on well log data.ipynb"

import pandas as pd
import numpy as np
import lasio
from pychangepoints import cython_pelt, algo_changepoints

Loading and cleaning data

well_log_las = lasio.read('./data/2120913D.las')
well_log_df = well_log_las.df()

list_var=['GR','NPOR','RHOZ','PEFZ']
well_log_data = well_log_df[list_var]

well_log_data = well_log_data[well_log_data>-1]
well_log_data.dropna(inplace=True)
well_log_data = well_log_data[well_log_data.index>6542]

Segmentation with PELT for multivariate time series.

penalty_value = 5
min_segment_size = 20
model = 'mbic_mean'
list_cpts, nb_cpts = algo_changepoints.pelt_multiple(well_log_data,penalty_value, min_segment_size, model)

Documentation

PELT

Inputs:

• data: data as pandas dataframe,
• penalty_value: double, penalty value, (BIC penalty),
• min_segment_size: int, minimum segment size,
• model: str, statistical model (see cost_function.pxd).

Outputs:

• list_cpts: array of the postion of the change-points,
• nb_cpts: number of chnagepoint.

Usage:

from pychangepoints import algo_changepoints

penalty_value = 5
min_segment_size = 20
model = 'mbic_mean'
list_cpts, nb_cpts = algo_changepoints.pelt(data, penalty_value, min_segment_size, model)

For the multivariate case, call pelt_multiple:

from pychangepoints import algo_changepoints

penalty_value = 5
min_segment_size = 20
model = 'mbic_mean'
list_cpts, nb_cpts = algo_changepoints.pelt_multiple(data, penalty_value, min_segment_size, model)

NP PELT

Inputs:

• data: as pandas dataframe,
• penalty_value: double, penalty value, (BIC penalty),
• min_segment_size: int, minimum segment size,
• nquantiles: int, number of quantiles,
• method: str, cost functioon (optional, only one implemented).

Outputs:

• list_cpts, array of the postion of the change-points
• nb_cpts, number of chnagepoint.

Usage:

from pychangepoints import algo_changepoints

penalty_value = 5
min_segment_size = 20
model = 'mbic_nonparametric_ed'
nquantiles = 10
list_cpts, nb_cpts = algo_changepoints.np_pelt(data, penalty_value, min_segment_size, nquantiles, method = model)

Binary Segmentation

Inputs:

• data: pandas dataframe,
• Q: int, number of changepoints,
• min_segment_size: int, minimum segment size,
• statistical model, str (see cost_function.pxd).

Output:

• array of the postion of the change-points,

Usage:

from pychangepoints import algo_changepoints
Q = 5
min_segment_size = 20
model = 'mbic_mean'
list_cpts = algo_changepoints.binseg(data, Q, minseg, method)

For the multivariate case, call binseg_multiple:

from pychangepoints import algo_changepoints
Q = 5
min_segment_size = 20
model = 'mbic_mean'
list_cpts = algo_changepoints.binseg_multiple(data, Q, minseg, method)

Segmentation neighborhood

Inputs:

• data: pandas dataframe,
• Q: int, number of changepoints,
• statistical model, str (see cost_function.pxd).

Output:

• array of the postion of the change-points,

Usage:

from pychangepoints import algo_changepoints
Q = 5
model = 'mbic_mean'
list_cpts = algo_changepoints.segneigh(data, Q, method)

For the multivariate case, call segneigh_multiple:

from pychangepoints import algo_changepoints
Q = 5
model = 'mbic_mean'
list_cpts = algo_changepoints.segneigh_multiple(data, Q, method)