carsonfarmer / predicure

A sketch-based semi-Markov learner with approximate duration estimation.

predicure

A sketch-based semi-Markov learner with approximate duration estimation.

Overview

This module implements an on-line (streaming) data structure for 'learning' semi-Markov models. A Markov model is a stochastic model used to model randomly changing systems where it is assumed that future states depend only on the current state, and not on the events (states) that occurred before it [wikipedia]. A semi-Markov process is one in which the probability of there being a change in state additionally depends on the amount of time that has elapsed since entry into the current state.

The SemiMarkov data structure within this module is designed to take in state change events (state 1 -> state 2) and their corresponding holding times (duration), and update a simple semi-Markov model. As each new event is added to the data structure, the data structure...

1. Updates state 1 duration distribution, and
2. Updates the state 1 transition counter and set of possible transition states state 2

There are tons of resources for computing Markov models, a quick Google will find lots of implementations.

Installation

predicure has not yet been uploaded to PyPi, as we are currently at the 'pre-release' stage*. Having said that you should be able to install it via pip directly from the GitHub repository with:

pip install git+git://github.com/carsonfarmer/predicure.git

You can also install predicure by cloning the GitHub repository and using the setup script:

git clone https://github.com/carsonfarmer/predicure.git
cd addc
python setup.py install

Note that predicure is written for Python 3 only. While it may work in earlier versions of Python, no attempt has been made to make it Python 2.x compatible.

* This means the API is not set, and subject to crazy changes at any time!

Testing

predicure comes with a comprehensive a very basic range of tests. To run the tests, you can use py.test (maybe also nosetests?), which can be installed via pip using the recommended.txt file (note, this will also install some other stuff (numpy, scipy, and matplotlib) which are all great and useful for tests and examples):

pip install -r recommended.txt
py.test predicure

Features

In the following examples we use the random module to generate data.

from predicure import SemiMarkov
import random

Basics

The simplest way to use a SemiMarkov data-structure is to initialize one and then update it with data points (via the update method). In this first example... and add them to an SemiMarkov object:

ac = SemiMarkov().batch(data)  # Add paths all at once...

We can then add additional data, and start to query the data-structure. As transitions are added, the data-structure responds and updates accordingly.

To illustrate the use of this data-structure, here is an example plot using the cluster_data from above:

import matplotlib.pyplot as plt

plt.figure()
mod.plot()
plt.show()

License

Copyright © 2016, Carson J. Q. Farmer
Licensed under the BSD-2 License.