4.5-1 Fixed memory leak in av_basis which is called by AvBasis.wst

4.6 Changed PsiJ. Added new option verbose, so that the information messages
	that previously were always printed out can now be suppressed.
	Also, changed default value of OPLENGTH to be 100000

    Changed wvrelease. Changed the previous two uses of "cat" to now use "message", which can be suppressed. This helps the package knitr

    Changed draw.default. Added a type argument.

    Changed nlevels to nlevelsWT. This was because nlevels clashed with
	a function of the same name in another package to do with factor
	levels.

4.6.1  Merged the contents of the cthresh package into wavethresh. Keeping
    the two as separate packages was causing big problems.

4.6.2  Minor bugfixes.

	In CWCV...

	Changed the argument "interptype" in "CWCV" to "noise". Sometimes
	the "normal" argument causes the routine not to terminate.


	In WaveletCV...

	Two function calls to "threshold" needed the argument
	policy="universal" added. This is because they were using
	the "sure" policy by default and the routine required a
	universal threshold. Note, people would have probably been
	using CWCV so this error would not have been noticed much.
	The bug comes to the fore when one changes the threshold type
	from the default "soft" to "hard", as SURE thresholding
	can't be computed with the latter.

	In rsswav...

	Function was returning a multi-argument, which is no longer
	allowed, changed so that a list is returned.

	In PsiJ and PsiJmati...

	The default value of the OPLENGTH argument has been increased to 10^7

	In PsiJ,ipndacw,rmget...

	Various matrices and objects get stored in an environment (WTEnv)
	now. So, invocation of one of these functions *the first time*
	will result in the object being computed from scratch and stored.
	Repeat invocation will used the stored objects saving computing time.
	This is different to previous versions of WaveThresh. In previous
	versions the stored objects got stored in your home R directory
	(ok .GlobalEnv) without your permission and without your knowledge
	(unless you looked) - but they were useful. The previous version
	though did have the advantage that the stored objects persisted across
	different sessions of R, or could be used if you attached another
	directory which had the stored objects in. 

	I'd like to see how things go with the current version. If
	continued recomputing is a pain then let me know and I'll think
	up some different model.

4.6.3	Make some corrections after running valgrind. See comments in code
	for details - search on Valgrind

	In function evalF, SFDE6, SFDE5, PLDE2 (these latter functions
	were also modified to catch and report on failures of memory
	allocation and also to properly free memory once it had been used). 	

4.6.4	The h_n quadrature mirror file coefficients for the (little-used)
	Littlewood-Paley wavelet have been changed so that they
	sum to \sqrt(2) and their sum of squares is 1. This is to bring
	them in line with the Daubechies compactly supported wavelets

4.6.5	Changed CWaveletCV to count the number of iterations in the
	main iterative process (using the counter iterations) and to
	check that it is always less than the supplied argument
	maxits. This is so that calls to this routine (e.g. from the
	much used threshold.wd using the "cv" policy do not run on without
	limit). The R code CWCV and threshold.wd have also been changed
	to take account of new maxits and cvmaxits arguments and associated
	help pages changed.

4.6.6	Bugfix in plot.mwd found by Sebastian Muller. This was caused
	by a reference to nlevels() which is not a WaveThresh function
	anymore (its name got changed to nlevelsWT() in version 4.6)
	This call was accidentally left over. 
	Also took opportunity to bring the DESCRIPTION file up to date.

4.6.7	Remove all references to the tsp function. This was a compatibility
	fix in R to relate to S, but now causes problems.

4.6.8	Added Coiflets to the filter.select function and updated help
	file (thanks to Anestis Antoniadis. I could have sworn I
	put these in before, but apparently not).

	Fixed bug. This bug only manifested itself in Windows code.
	In Best1D cols an operation occurs which compares all wavelet
	packets to the groups vector, to identify which packets are
	similar enough to the groups vector for further consideration.
	The first wavelet packet, though, is always a constant vector
	and hence the correlation is always zero (unless the groups
	vector was itself constant, but then this is not an interesting
	case). Now, we set this first value equal to zero. Previously,
	the correlation was computed but the denominator involves calculating
	the standard deviation of the constant vector, which is zero, and
	this causes a division by zero warning and returns and NA which
	can't then be handled. Now setting the correlation for this entry
	to be zero is correct and bypasses the problem.
	
4.6.9	Added sanity checks to the arguments of makegrid. The length of the
	vector arguments t and y should be the same. However, the function
	did not check this and erroneous NA values appear when the function
	was called on vectors of different lengths. The function now checks
	and stops if the vectors are of different lengths. Additionally,
	checking on the length of the gridn number, which should be dyadic
	is also performed, stopping the function if it is not.

	Changed maintainer email address to g.nason@imperial.ac.uk

	Added routines to register routines (src/wavethresh_init.c)

	Bugfix on line 5494 of functions.c. Line band1- changed to band1--,
	as band1 needs to be successively decrememented (as with band2++ later).
	This is in the computec C function which gets called by irregwd

4.7.0 	Changed the few references from Sfloat to double and Sint to int

	Changed lines in function.r that were comparing the output of class()
	to string to make use of the inherit function instead.

4.7.1	Inserted C prototypes (in response to package update request from R
	team)

4.7.2	Corrected typing of wpCmnv function which was wrong in
	wavethresh_init.c
