http://homepages.inf.ed.ac.uk/lzhang10/maxent.html
This package provides a (Conditional) Maximum Entropy Modeling Toolkit for Python and C++. The library is written in ISO C++ and has been tested under GNU/Linux, FreeBSD/NetBSD and Win32 (Cygwin and http://www.mingw.org). There is also a python extension module (maxent module) available.
For more information, please have a look at the file manual.pdf in doc/ directory. The PDF document talks about the toolkit at length.
See file INSTALL for detail description of building the maxent package on unix platforms. Instruction on building under win32 environment is covered in the PDF manual in doc/.
python/ directory contains source code for python binding of the toolkit. The detail instruction on building the python extension is given in python/README.
This software is freeware and is released under LGPL. see LICENSE file for more information.
The adoption of LGPL is in accord with the license of java maxent project: http://maxent.sf.net, from which the toolkit was derived. LGPL makes it easier to share source code, as well as new ideas, between both projects.
Since binary feature and integer feature are more common than high precision real feature, all feature values are represented as float (4 bytes) rather than double (8 bytes) in order to save memory. If you need to specify double feature values you can find all floats in source code and replace them with double (require more memory).
Pull requests are welcome.