segmented 0.0.1

segmented: A toolbox for segmented regression.

Segmented

segmented is a Python toolbox for segmented regression.

Installation

segmented is written for Python 3.7+. To install...

Features

• 0.0.1
  • single change point
  • identity link functions
  • single predictor
• 0.0.2
  • single, parametric change point
  • identity link functions
  • single predictor

• = 0.0.3

  • user-specified number of segments (>2)
  • link functions
  • etc.

How to Use segmented

import segmented as sgmt
import pandas as pd

data = pd.read_csv('mydata.csv')

###############################################################################
# 2 segment model (1 change point)

# specify 2 model specifications: pre- and post-changepoint
model = sgmt.segment(models=['y~x', 'y~x'], data=data)

# specify a single model specification that will be used for each segment
model = sgmt.segment(models=['y~x'], num_segments=2, data=data)

# might need to provide some initial guesses at to-be-estimated parameters
model = sgmt.segment(models=['y~1', 'y~x'], x0=.5, data=data)

# might provide GLM capabilities
model = sgmt.segment(models=['y~x'], family=['binomial'], num_segments=2, data=data)
model = sgmt.segment(models=['y~x'], family=['binomial'], num_segments=2, data=data)

# estimate parameters
model.fit()
# check out estimates
print(model.summary())

OR

###############################################################################
# 2 segments, each segments has an intercept and slope associated with x

# the changepoint is a single value of x
model = sgmt.segment(models='y~x', changepoint='~1', num_segments=2, data=data)
# is equivalent to:
model = sgmt.segment(models='y~x', num_segments=2, data=data)

# 2 segments, each segments has an intercept and slope associated with x
# changepoint is parametric, with it's own intercept and slope associated with z
model = sgmt.segment(models='y~x', changepoint='~z', num_segments=2, data=data)

model = sgmt.segment(models=['y~x','y~x','y~x'], changepoint=['~z','~w'], data=data)

References

• TBD