dgim 0.2.0

Features

• Estimation of the number of “True” statements in the last N element of a boolean stream

• Low memory footprint.

• Tunable error rate (the lower the error rate, the higher the memory footprint)

Applications

When processing large streams of data such as clicks streams, server logs, financial streams. It is often necessary to maintain statistics about the N latest elements. If N is big or if you have many streams to process, it is not possible to store the N latest elements.

In such situations, if the processed stream is made of boolean, the DGIM algorithm can help you estimate the number of True statements in the last elements.

For instance, if the stream is made of server logs, DGIM algorithm can estimate the proportion of visits that come from search engines. (as opposed to direct access, or access through paid search)

Installation

At the command line:

$ pip install dgim

Usage

Sample code:

from dgim import Dgim
dgim = Dgim(N=32, error_rate=0.1)
for i in range(100):
    dgim.update(True)
dgim_result = dgim.get_count() # 30 (exact result is 32)

Documentation

https://dgim.readthedocs.org.

License

The project is licensed under the BSD license.

Authors

• Simon Dollé <simon.dolle@gmail.com>

• No other contributor yet. Why not joining?

How to contribute

1. Check for open issues or open a fresh issue to start a discussion around a feature idea or a bug.

2. Fork the repository on GitHub to start making your changes to the master branch (or branch off of it).

3. Write a test which shows that the bug was fixed or that the feature works as expected.

4. Send a pull request and bug the maintainer until it gets merged and published. :) Make sure to add yourself to AUTHORS.

References

• Datar, Mayur, et al. “Maintaining stream statistics over sliding windows.” SIAM Journal on Computing 31.6 (2002): 1794-1813.

• Rajaraman, Anand, and Jeffrey David Ullman. Mining of massive datasets. Cambridge University Press, 2011. Chapter 4. http://infolab.stanford.edu/~ullman/mmds/ch4.pdf

• Mining of Massive Datasets Coursera MOOC: http://infolab.stanford.edu/~ullman/mmds/ch4.pdf

History

0.2.0 (2015-01-05)

• Improved documentation

• Make most methods and attribute private.

0.1.0 (2015-01-04)

• First release on PyPI.