v1.3.3, 19/8/2014
 - Correct use of kernel in non-parametric regression
 - Added attributes to BootstrapResult
 - Added efficient binning in Cythong and Python
 - If using Cython, linear binning is used, which is better for small bin numbers
 - Using rfft to save time (computes half the values)
v1.3.1, 8/8/2014
 - Removed necessity of calling fit
 - Added warning to main text
 - Small FFT optimization
v1.3.0, 8/8/2014
 - Removed unnecessary type conversion
 - Rewrote the non-parametric fitting
 - Rewrote parametric fitting
 - The interface should be now uniform across the board, but incompatible with previous versions
v1.2.5, 6/8/2014
 - Using 'out' rather than 'output' for in-place operations to conform to numpy convention
 - Added a kernels.Kernel base-class implementing cdf, pm1, pm2, fft and dct in term of the pdf
 - Corrected all the kernels, both Python and Cython
 - Added and restructure kernel tests
 - Restructured KDE tests.
 - Moved TransformKDE to the kde_methods
 - Updated tutorial
 - Added decorators for 1 argument numpy transformations (i.e. functions working 
   on arrays and returning an array of same shape as the input)
v1.2.4, 30/7/2014
 - Corrected KDE code to pass tests
 - All tests pass for Python2 and Python3
 - Corrected loader for python 3.4
v1.2.2, 30/7/2014
 - Corrected distribution + version + CHANGES
v1.2.0, 30/7/2014
 - Corrected in kernel_smoothing the use of atleast_?d to allow non-ndarray arguments
 - Changed interface to select KDE methods to use objects
v1.1.17, 29/7/2013
 - Corrected kde.KDE1D initialization and weigth handling
 - Corrected CSV output of fitting
 - Corrected the kde.botev_bandwidth to work with small input sizes
v1.1.16, 28/5/2013
 - Remove need for cython in any module (was partial)
 - PEP8 compliance
v1.1.15, 28/5/2013
 - Changed description
 - Doc now hosted at readthedocs.org
 - Hosting package at PyPi
v1.1.14, 27/5/2013
 - Unifying code base using a compatibility module
 - sharedmem.py is now the same for Python2 and Python3 (and working with python3!)
 - cleaned and corrected CSV output code for python 2 and python 3
v1.1.13, 21/5/2013
 - Removed reference to npy_float128. Seems it doesn't work on 32 bits machines.
 - Adapted to 32 bits machines
 - Changed integration from basic to Simpson's method
 - If cython is not present (on not wanted!), kernels exist also as pure python
   functions
 - Free switching between pure python and cython kernels
 - Cython 1.19 is not really needed, so downgrading the need to 1.17
 - Moved Cython dependencies into an "extra" with name 'Cython'
v1.1.12, 15/5/2013
  - Added transformed KDE class, to compute KDE using variable change
v1.1.11, 27/2/2013
 - Correct CDF of tricube kernel (error inducing a 1e-4 error on CDF computation)
 - Added tests for kernels
 - Added new interface for 1D local polynomial -- compatible with previous one!
 - User can specify their own design matrix to allow for non-polynomial basis
v1.1.10, 26/2/2013
 - Added Epanechnikov kernel
 - plot_residual_tests now supports nD regression
 - corrected kde in case no method is specified
 - corrected grid_eval (now works)
 - corrected weights application
v1.1.9, 18/2/2013
 - Added tests for the various modules
 - Corrected bug in sharedmem for Python 3
 - Now using print_function and relative_import futures to minimise changes between Python 2 and 3
v1.1.8, 29/1/2013
 - Corrected Cython import error in kernels.py
 - Added "from __future__ import division" in all numerical module
v1.1.7, 28/1/2013
 - Corrected installation bug
 - Started tutorial on non-parametric fitting
 - Corrected a few bugs in Spatial Averaging class
v1.1.6, 27/1/2013
 - Corrected a few issues with documentation to allow for PDF generation
 - Changed the __init__.py to remove references to pylab, so it works without GUI
 - Corrected grid code to account for weights and lambdas
v1.1.5, 25/1/2013
 - Added CDT and FFT-based KDE estimation
 - Added TriCube kernel for demonstration
v1.1.4, 23/1/2013
 - Refactored 1D KDE estimation to support multiple methods.
 - Refactored documentation to reflect this
v1.1.3, 20/1/2013
 - Added nD local polynomial regression
 - Added 1D bounded KDE estimation
v1.1.2, 11/1/2013
 - Added 1D local polynomial regression
v1.1.0, 8/1/2013
 - Curve fitting changed to match non-parametric interface
 - Unified bootstrap
v1.0.23, 6/1/2012
 - Adapted the CurveFitting to fit the new bootstrap
v1.0.22, 6/1/2013
 - Changed interface to add user functions, using objects now
 - Added start of documentation
 - Changed function and residual module loader to avoir name conflicts
 - Corrected the loading of user-defined functions
v1.0.21, 5/1/2013
 - Improved description
 - Changed dependancy from setuptools to distribute because of a problem with setuptools 0.6
v1.0.20, 5/1/2013
 - Made non-parametric bootstrap method parrallel
 - LocalLinear1D is done in Cython now
 - Use LocalLinear1D for residual analyses rather than SpatialAverage
 - Changed bootstrap method to be more general.
 - Changed the functions and residuals modules to read from the PYQTFIT_PATH environment variable the list of extra folders to analyse
 - Added Local Linar Kernel Regression method
v1.0.19, 10/12/2012 - Corrected dependency specifications.
v1.0.14, 4/12/2012 - Added version numbers for the requirements
v1.0.10, 4/12/2012 - First release distributed with pip
v1.0.9, 4/12/2012 - Last release without pip

