peewit intends to facilitate programming, running and result examination of machine learning experiments. It does not provide any sort of ML data types, engine, or interfaces, but it give some assistance in avoiding a mess. The user is to decompose the experimetnal code into named nodes that correpsond to aspects or dimension of an experiment. Since this breakdown is readable for machines too, simple services can be precoded that help to keep track of the experimental dimensions. Further, it puts some regime on output files, helps to recover former versions, and handles simple parallelization.
Be aware that the implementation still has deficiencies: it is weakly tested, many doc-string are missing, erroneous calls are not always caught by helpful error messages and there is much left for smaller and greater amendments. At least, the examples are decorated with many explicatory comments. We do not recommend peewit at the current state unless you are curious about it. Also, after release of version 0.7 we overworked the experiment-model and then rewrote the core modules.
The core modules depend on numpy only but for full functionality you further have to provide unison, ssh with agent, git, and matplotlib, as well as libsvm for the example project. Feel free to contact us for questions.
- Changes to previous version:
Other available revisons
Version Changelog Date 0.10
v-cube with side-cubes
May 7, 2014, 16:04:18 0.9
switched to python-3
February 11, 2013, 21:21:05 0.8
July 9, 2012, 23:32:20 0.7
November 4, 2011, 19:54:09 0.6
September 18, 2010, 16:22:01 0.5
August 18, 2010, 14:00:21 0.4
platform-independent path handling
July 21, 2010, 15:05:31 0.3
incremental e-tree definitions
May 12, 2010, 21:44:21 0.2
self-referential descent inputs
May 4, 2010, 13:30:36 0.1
services for arg-aggregations
April 26, 2010, 11:17:50 0.0
Initial Announcement on mloss.org.
April 23, 2010, 23:14:51
No one has posted any comments yet. Perhaps you'd like to be the first?
Leave a comment
You must be logged in to post comments.