Changes since version 1.2.4
	
	* Fixed a bug in predict.R, thanks to John Eric Humphries of U Chicago for the catch.

	* Replaced OpenMP C++ compiler check from the deprecated SUPPORT_OPENMP to _OPENMP

	* Included CITATION file to reflect JSS paper.

Changes since version 1.2.3

	* Fixed a bug in likelihood calculation when 'weights' isn't NULL. Thanks to Fan Yang of U Chicago for reporting.
	
	* Weights are now normalized such that their sum equal num obs.

Changes since version 1.2.2

	* Fixed a bug in AIC calculation
	
	* Fixed a requirement in predict for newdata to have response column

Changes since version 1.2.1
	
	* Made a change in mnlogit.R return field est.stat so that pre-processing time is returned.
	
	* Made a change in mnlogit.R return field est.stat so that pre-processing time is returned.

Changes since version 1.2.0
	
	* Fixed a bug in predict(). Thanks to Frank Zhe Wang of UNC for reporting.

	* Fixed a bug in predict() for handling choiceVar arg. 

	* Fixed a bug in mnlogit.R in getting correct order of names for choices. Thanks to Matthew Levinson for reporting.

	* Added droplevels() when converting response col to T/F in mnlogit.R

	* Added UseMethod() to allow index() to be called on mnlogit objects

	* Added argument 'what' in print.mnlogit()

	* Slightly changed wording of output in print.model.size 

	* Minor correction in comments in file mnlogit/src/hessian.h

Changes since version 1.1.1

	* Made choiceVar arg in mnlogit() optional when data is a mlogit.data obj.

	* Added arg alt.subset in mnlogit() call to estimate on a subset of alternatives.

	* Added arg start in mnlogit() call to accept an initial guess for coeffs.
 
	* Added hypothesis tests - 3 MLE test and one test for IIA.   
 
	* Added Several methods for mnlogit objects to match mlogit methods.
	
	* Residual calculation now matches mlogit.
	
	* The output of newtonRaphson() has been slightly changed for residual field.

	* mnlogit() returns an object with slightly different fields, matching mlogit.

	* formulaParser() now relies on the Formula package for manipulating formulas.

	* Fixed a bug in newtonRaphson() related to failed linesearch situations.

Changes since version 1.1.0

	* Fixed a critical bug in calling compiled code with incorrect matrix args.

Changes since version 1.0.1
	
	* Changed predict function using contributed code.

	* Included weights option in mnlogit() call.

	* Gradient norm for stopping now computed wrt inverse Hessian.

	* Fixed a bug with the lowest allowed step size during line search.

	* Changed .C() with DUP=FALSE to .Call() to meet CRAN requirements.

Changes since version 1.0.0
	
	* Inserted include guards for all openMP commands.

