[MRG] Add more multiclass datasets

[mathurinm]

sklearn.datasets.load_svmlight_file fails for this protein dataset, because it stores values as .50 instead of 0.50

Reproduce with a dummy_svmlight file containing a single line:
0 21:1.00 42:1.00 63: .50
and run load_svmlight_file("dummy_svmlight")

fix it by replacing .50 with 0.50

will have to dig a bit into load_svmlight to see how to fix

[codecov-io]

Codecov Report

Merging #6 into master will not change coverage.
The diff coverage is n/a.

@@           Coverage Diff           @@
##           master       #6   +/-   ##
=======================================
  Coverage   89.58%   89.58%           
=======================================
  Files           3        3           
  Lines          96       96           
  Branches       14       14           
=======================================
  Hits           86       86           
  Misses          5        5           
  Partials        5        5           

Impacted Files	Coverage Δ
libsvmdata/datasets.py	86.84% <ø> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update bf51fab...4243523. Read the comment docs.

[mathurinm]

In the cython code at https://github.com/scikit-learn/scikit-learn/blob/0fb307bf39bbdacd6ed713c00724f8f871d60370/sklearn/datasets/_svmlight_format_fast.pyx#L68
there is

target, features = line_parts[0], line_parts[1:]

which in this case yields

['21:1.00', '42:1.00', '63:', '.50']

a potential fix is to perform line = line.replace(': .', ':0.')
will have to do benchmarks to see if it harms performance. Given that it's for a single dataset, it may not be worth it so far.

@QB3 here's the explanation for the issue you encountered.

[agramfort]

sklearn code should be fixed :(

…