Codecov Report

Merging #619 (f45a2b0) into master (b8f2717) will decrease coverage by 0.07%.
The diff coverage is 82.50%.

@@            Coverage Diff             @@
##           master     #619      +/-   ##
==========================================
- Coverage   90.43%   90.36%   -0.07%     
==========================================
  Files          20       20              
  Lines        3564     3601      +37     
==========================================
+ Hits         3223     3254      +31     
- Misses        341      347       +6     

Okay, so I looked through the standard -- For unique_inverse and unique_all, it doesn't make sufficient sense to implement them, because there's "no better algorithm than dense", please let me know if I'm missing something. The other two, it definitely makes sense to implement.

Okay, so I looked through the standard -- For unique_inverse and unique_all, it doesn't make sufficient sense to implement them, because there's "no better algorithm than dense", please let me know if I'm missing something. The other two, it definitely makes sense to implement.

For unique_all I agree as indices component requires finding the first occurrence of the fill value, I can make it throw an exception saying that it isn't supported by COO.
For unique_inverse I construct inverse as a DOK and only plug non-fill values so it isn't a dense algorithm I think - but is inverse even useful here?

I removed unique_all and unique_inverse functions (updated PR's description).