Author: Zhou Fan Version: 1.0.2 Description: This library is a Python/C++ implementation of the algorithms described in Fan, Z. and Mackey, L. ``An Empirical Bayesian Analysis of Simultaneous Changepoints in Multiple Data Sequences''.

INSTALL

This package requires a working installation of Python 2.7, numpy, and matplotlib. From a terminal, cd to the pylib subdirectory and execute

python setup.py build python setup.py install

To test your installation, cd to the pylib/test subdirectory and execute

python test.py

This will test all of the implemented likelihood model and plot simulated data with detected changepoints.

To install to a local directory (e.g. if you do not have administrative access to the default install location), cd to the pylib subdirectory and execute

python setup.py build python setup.py install --prefix=

where should be replaced by a local installation directory. Ensure that /lib/python2.7/site-packages is added to your PYTHONPATH environment variable (e.g. by adding the line

export PYTHONPATH=/lib/python2.7/site-packages:$PYTHONPATH

to your .bashrc or equivalent file).

The installation assumes that the default C++ compiler used by Python has C++-11 support. The compiler may be manually set within the setup.py file by adding the lines

import os os.environ["CXX"] = <YOUR-C++-COMPILER>

USAGE

To read a CSV-formatted data matrix 'X.csv' (with rows indexing sequences and columns indexing sequential positions/time points) into Python:

import numpy X = numpy.loadtxt('X.csv', delimiter=',') (J,T) = X.shape

To draw 200 samples of changepoint locations using MCMC and estimate changepoint model parameters using MCEM (assuming a Gaussian observation model with changing mean and fixed variance):

import BASIC_changepoint as BASIC [samples, model_params, pi_q, q_vals] = BASIC.MCMC_sample(X, 'normal_mean', sample_iters=200, MCEM_schedule=[10,20,40,60,100])

To obtain a posterior mode point-estimate of changepoints, using the MCMC-estimated posterior mean rounded to 0--1 as an initial guess:

chg_probs = BASIC.marginal_change_probs(samples[100:], J, T) mode = BASIC.compute_posterior_mode(X, 'normal_mean', model_params, pi_q, q_vals, Z=(chg_probs>0.5))

To plot the first four sequences in X with their detected changepoints:

fig = BASIC.plot_changes(X, mode, seqs=[0,1,2,3]) fig.show()

To save the MCMC sampling output to a (compressed) file 'MCMC_out.pkl' on disk, and to read it back from disk:

import cPickle cPickle.dump([samples, model_params, pi_q, q_vals], open('MCMC_out.pkl','w')) [samples, model_params, pi_q, q_vals] = cPickle.load(open('MCMC_out.pkl','r'))

To convert the posterior mode point-estimate to a binary 0--1 matrix of changepoint indicator variables and save it to a CSV-formatted text file 'Z_mode.csv':

Z = BASIC.change_dict_to_Z_matrix(mode, J, T) numpy.savetxt('Z_mode.csv', Z, fmt='%d', delimiter=',')

For further documentation on MCMC_sample, compute_posterior_mode, and other functions in the BASIC library:

help(BASIC)