Changes in Version 1.1 (9th August 2005)
-------------------------------------------------------------------
Cross validation root mean squared standardised error
is now reported correctly when 'leave final 20 percent out'
is selected as the cross-validation option

Uncertainty analysis can now be calculated without doing sensitivity analysis
(added 'uncertainty analysis' checkbox to options page)
Old v1.0 Project files will not work because of new additions to the project file.

fixed bug which occurred when setting input column to integer
(subsequent rescaling can now take place)

User is now asked to confirm if project has been modified between saving and quitting GEM-SA. 

Changed order of tabs in project dialog. Files now come before options.

Allowed option to set number of inputs directly from the inputs file
(this is why it makes sense to specify the files first)

Output Sigma-squared instead of precision as a measure of emulator accuracy,
to allow for easier interpretation. (sigmasquared = 1/precision)

Total effects are now consistent with the main effects (rather than scaled to add to 1).
Due to over-counting, total effects will now sum to more than 100% unless the output is
additive

Cross validation outputs now include the largest standardised error. This can indicate where,
in the input space, the Gaussian process assumptions are most questionable, and where more
runs may be required.

The code inputs file can now include an optional header line with a name for each input (column).
This header line is optional and any names specified in the project override (and will not be replaced
by) any header names. If any names are not specified in the project,
the header names will be used. If no header names are included in the input file,
the defaults x1, x2, ... are used instead.
A name canot contain spaces in the header line, but can in the project file.

Means and variances (if inputs are normally distributed)
or min and max (if inputs are uniformly distributed) can be chosen to have default
values using the ranges in the data. For normal inputs, mean and variances are set
so that the range of values in the data cover mean +- 2 standard deviations.
For uniformly distributed inputs, the min and max are the min and max values of the data.

Uniform distributions with min and max other than the default values (from the data) can
now be selected, and I added to the uncertainty options: Some inputs know, others uniform.

A bug in the stand-alone fortran version emulator.f,
which meant the standard deviations of predictions were wrong,
has been fixed.

Corrections to the prediction routines when some inputs are known

Improvements to the maximisation routine which selects the emulator roughness parameters

An output file with multiple columns, corresponding to multiple code outputs.
GEM-SA can only emulate 1 output at a time. The single output to emulate is specified in the
options tab. 

Values in the printed report and on the output summary page refer to
the unstandardised output scale.