A=GROUP(I,[G]) returns a group object initialized with
a cell array of algorithms I and an optional grouping type G.
This is used to collect a group of algorithms.
Examples: group({svm knn c45}),
get_mean(train(cv(group({svm knn })),train(toy)))
Hyperparameter
group='all' This parameter stores grouping type:
'one_for_each', 'all' or 'separate' (default='all').
When results of training and testing are
output by a (grouped) algorithm,
they are stored in the same group type given
here. This is important for objects like chain,
and get_mean which do not deal with data objects
separately.
GROUP TYPE
'all': each item is passed independently and separately
to all training objects, e.g
d=gen(toy); a=group({knn svm});
train(a,group({d d},'separate'))
gives 4 outputs.
'one_for_each': the n^th data item is passed to the
n^th training object in an group object, e.g
d=gen(toy);g=group({d d},'one_for_each');
a=group({knn svm}); train(a,g)
gives 2 outputs.
'all': passes all the data objects to a
single training object, e.g
d=gen(toy); a=group({knn svm}); r=train(a,d);
train(get_mean,group(r,'all'))
gives 1 ouput.
[NOTE: It is also possible to use the transpose operator with
group, e.g: r=group({ {svm svm} {knn knn} })' will give
the same result as group({ {svm knn} {svm knn} })]