Multiclass Support

[khdlr]

The introduction of multiclass support would be great

potential overlap with [TODO]

requirements:

• Training data/vectors attribute label has numerical values, which can be > 1

preprocessing:

• mask as uint8 with label ids as values

data output:

• class labels as uint8 as uint8 with label ids as values
• class probabilities
  • option 1: n bands with p values, n=number of classes (my preferred option)
  • option 2: 1 band, p value of majority class

numbers output:

• class metrics need to be adapted to multiclass (precision, recall, f-score)
• addition of new output: e.g. error matrix

[khdlr]

This is essentially ready to merge at #7