Peter M. Groen / openbr

Resolved conflicts:
	app/br/br.cpp
	openbr/core/bee.cpp
	openbr/plugins/output.cpp

resolved conflicts:
	app/br/br.cpp
	openbr/core/bee.cpp
	openbr/core/classify.cpp
	openbr/core/cluster.cpp
	openbr/core/eval.h
	openbr/openbr.cpp
	openbr/openbr.h
	openbr/plugins/algorithms.cpp
	openbr/plugins/independent.cpp
	openbr/plugins/output.cpp
	openbr/plugins/svm.cpp

Common::seedRNG uses a static variable to ensure that the RNG is seeded exactly
once per run of OpenBR. In spite of this, seedRNG was called from several
places, although the function would have an effect exactly once.

This commit
adds thread safety to seedRNG (previously the use of a static variable was
unsafe, although it could only cause a rather minor problem). More importantly
call seedRNG only from Context::initialize. If it is desirable to call seedRNG
more than once, changes will need to be made to support doing that in a
meaningful way.

Remove variables related to subsampling training data from Transform, move them
to a new transform called DownsampleTraining, and perform subsampling in that
transform, rather than as part of IndependentTransform (which depended on the
subsampling variables in Transform).

The prior approach was incompatible with using explicit class variable names,
rather than assuming a fixed label variable. This is because the actual
downsampling was carried out as part of IndependentTransform, which is not a
visible part of the algorithm grammar. Removing this logic from
Independent/Transform is generally desirable if only to avoid cluttering
Transform with member variables that are only used some of the time, and used
in a (fairly) restrictive way.

The current approach of makring DownsampleTraining as an indepent transform
is still limited since the downsample logic (still) cannot be applied without
the split used in Independent, and also DownsampleTraining can only be
implemented as a wrapper for another transform (because the argument to train
is const, and performing the donwsample logic in project would also impact the
testing case (since we have no way to exclude a transform used in training from
use in testing, for a fixed algorithm)).

Replace direct use of "Label" by most supervised learning algorithms with
inputVariable br properties (default value "Label")

Change default label name from Subjet to Label (since label is a more general
term).
Use different default variable names for classification (label), regression
(regressor/regressand), and clustering (ClusterID)
Update some (far from all) transforms to accept arguments specifying their
input/output variables.
Update eval classification to optionally take target variable names as
arguments

These names mirroring File functionality.
Fixed SVMDistance
dropped unnecessary second relabel in LDATransform::train

Remove global label/subject lookup table
Consistently use "Subject" rather than "Label", subject is assumed to be
convertable to QString. When desirable, map discrete subject values to ints.

For classifiers such as svm that require numeric labels, generate a string->int
mapping for the training data, and store it (local to the transform).

Utility functions for collecting all values of a given property (on a template
list), and mapping discrete property values to 0-based integers

Some outstanding issues include use of label/subject in matrix output

Remove subject/label methods from the API, replace some methods with general
methods taking a property name as an argument.

This breaks quite a few things