Add SinkhornVectorizer for larger distributions

[codecov-commenter]

Codecov Report

Merging #69 (c72a1ea) into master (7d9c5ca) will decrease coverage by 23.64%. The diff coverage is 73.18%.

@@             Coverage Diff             @@
##           master      #69       +/-   ##
===========================================
- Coverage   89.33%   65.69%   -23.65%     
===========================================
  Files          19       19               
  Lines        3161     3437      +276     
  Branches      594        0      -594     
===========================================
- Hits         2824     2258      -566     
- Misses        287     1179      +892     
+ Partials       50        0       -50     

Impacted Files	Coverage Δ
vectorizers/token_cooccurrence_vectorizer.py	51.84% <ø> (-37.16%)	:arrow_down:
vectorizers/linear_optimal_transport.py	64.63% <69.69%> (-23.96%)	:arrow_down:
vectorizers/__init__.py	100.00% <100.00%> (ø)
vectorizers/_version.py	100.00% <100.00%> (ø)
vectorizers/tests/test_common.py	99.82% <100.00%> (+0.01%)	:arrow_up:
vectorizers/coo_utils.py	7.77% <0.00%> (-83.18%)	:arrow_down:
vectorizers/distances.py	8.02% <0.00%> (-68.10%)	:arrow_down:
vectorizers/_window_kernels.py	26.08% <0.00%> (-53.92%)	:arrow_down:
vectorizers/transformers.py	63.25% <0.00%> (-30.23%)	:arrow_down:
... and 7 more

---

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 7d9c5ca...c72a1ea. Read the comment docs.

[jc-healy]

Hey Leland, this looks pretty solid. The build errors all seem to be around the two new sinkhorn tests. From the Travis logs it looks like it kicks in when we turn off numba (NUMBA_DISABLE_JIT=1). I'm happy to merge one the tests pass.

[lmcinnes]

1. You are supposed to be on vacation.
2. Yeah, there's some weirdness, but I think I can fix it in due course.