Lsqfit

=======================
This is a modest change but includes some backwards incompatible changes in
the routine --- hence the increase to v7.0.

- nonlinear_fit now supports priors with log-normal and sqrt-normal
distributions provided parameter extend=True is set when it is called.
A parameter 'c', for example, in a parameter dictionary can be assigned a
log-normal prior by specifying a prior for 'logc' (or 'log(c)') in the fit
prior, rather than for 'c'. This is the only change required to switch from
'c' from a Gaussian prior to a log-normal prior; in particular the fit
function can be still be expressed in terms of 'c' (rather than 'logc') since
parameter dictionaries created by nonlinear_fit will have entries for both
'c' and 'logc'. sqrt-normal distributions are handled the same way, but
with, for example, 'sqrtc' or 'sqrt(c)' instead of 'logc'. This
functionality was available in earlier releases using the function decorator
transform_p, which is no longer needed and has been removed.This means that
fit.transformed_p is gone too --- just use fit.p.

- lsqfit.wavg no longer stores chi2, dof, etc as attributes of the
function. These are all attributes of the result returned by
wavg.

- debug=True checks for more errors. In particular it now looks for
stray GVars (not from the parameters) in fit functions. The only
way a GVar should get into a fit function is through the parameters.
Having other GVars there used to lead to a very obscure error message.
Running with debug=True gives a slightly more comprehensible message.

- fixed bug in nonlinear_fit.format which sometimes put stars in the wrong
place.

- modified nonlinear_fit.format to list extra entries generated when
using log-normal or sqrt-variables. So for a log-normal variable 'logc',
results for both 'logc' and 'c' will appear in the formatted output.

===========================
Changed to pip + distutils for installation. Fixed inconsistencies in
INSTALLATION.txt.

==========================
Going back to distutils instead of setuptools. The latter doesn't work well
with cython.

==========================
Fixed trivial error in MANIFEST.in which prevented building.

==========================

With this version gvar is no longer bundled with lsqfit. setuptools will
try to install gvar if it is absent; otherwise it can be installed directly
(pip install gvar). The undbundling facilitates the use of gvar by modules
unrelated to lsqfit. Except for the unbundling this version of lsqfit is the
same as the previous version and continues to run with the older (v6.0)
version of gvar. It also runs with the new version of gvar (v7.0), which
has some added functionality.

- setuptools is now the default installer, although distutils will be used
if setuptools is not installed.

- documentation loses gvar sections.

========================

This upgrade deals with some issues created by the most recent update to
numpy (v1.9.0). The verion number increases to 6 because a (very minor) part
of the gvar.powerseries interface had to change (and therefore by the rules of
semantic versioning one must increase the major verion number). BufferDicts
are also subtly changed, though in a way that shouldn't affect much code.
A couple of enhancements are included as well.

- gvar.BufferDicts are now derived from collections.OrderedDict. This is
a consequence of the numpy change, but it is probably a good idea anyway.
The interface is the same but pickling has changed. This means that
BufferDicts pickled with the old software cannot be unpickled with
the new. A very minimal module, oldbufferdict.py, is included in lsqfit
for converting old pickled data to the new format; see documentation
inside the file. Data stored using json is unaffected.

- The power series coefficents of gvar.powerseries.PowerSeries variable
p are now accessed via attribute c: p.c[0], p.c[1] etc. Formerly one
could access the coefficients using array notation --- p[0] for p.c[0],
etc --- but this no longer works with numpy. The old notation was a bad
idea in any case.

- gvar.fmt_errorbudget has a new option called verify. Set verify=True
to have the code check that the error budget is complete and has no
double counting of errors; a warning is issued if this is not the case.

- Added arctan2 to the list of functions that work with GVars,
using code from Matt Wingate. (NB, the documentation shows how to
create additional functions (eg, erf(x)) using gvar_function.)

- Obscure bug fix in gvar.ode.Integrator allows GVars as the
independent variables.

- Rearranged documentation now includes case studies. These are self
contained examples that are extensively annotated (and relatively simple).
There are only two at the moment but more are forthcoming.