Multilabel active learning support

[yangarbiter]

#64

Still need some discussions on interface for multilabel models.

Things to check:

• interface for multilabel model
• the DummyCLF for resolving conditions of some label might have only one class in training set
• test for multilabel model
• whether to create libact.models.multiclass and libact.models.singleclass

[codecov-io]

Current coverage is 85.86% (diff: 89.38%)

Merging #74 into master will decrease coverage by 0.67%

@@             master        #74   diff @@
==========================================
  Files            21         24     +3   
  Lines           676        679     +3   
  Methods           0          0          
  Messages          0          0          
  Branches          0          0          
==========================================
- Hits            585        583     -2   
- Misses           91         96     +5   
  Partials          0          0          

Powered by Codecov. Last update 2ef52ba...4701bd6

[iamyuanchung]

I think for the second issue, the problem is not about the design of interface but the algorithm itself - it's best for the algorithm to be capable of detecting new kinds of y \in {0, 1}^k.
I know for multi-label classification, there will be plenty of label combinations.
I will check how DummyCLF is working and comment more later.

[iamyuanchung]

For the last issue, what is the goal of libact.models.singleclass?