[REVIEW] Feature: pairwise edit distance for each string on a given nvstrings object

[kayush2O6]

This PR implements the edit_distance_matrix() function that computes pairwise distance on a list of strings(i.e. nvstrings object) and returns a distance matrix.
Recently, I was working on a strings related use-case and where I felt the need of this function...

Example:

import nvstrings, nvtext
s = nvstrings.to_device(["honda","hyundai","suzuki"])
n = nvtext.edit_distance_matrix(s)
print(n)
Out:    
[[0, 3, 6],
 [3, 0, 5],
 [6, 5, 0]]

cc @davidwendt

[codecov]

Codecov Report

Merging #2803 into branch-0.10 will not change coverage.
The diff coverage is n/a.

@@             Coverage Diff              @@
##           branch-0.10    #2803   +/-   ##
============================================
  Coverage        86.51%   86.51%           
============================================
  Files               48       48           
  Lines             9013     9013           
============================================
  Hits              7798     7798           
  Misses            1215     1215

---

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 b96073c...ff9eced. Read the comment docs.

[davidwendt]

Nice job Ayush. I like how you minimized the compute buffer too.

[kayush2O6]

Nice job Ayush. I like how you minimized the compute buffer too.

Thanks David!

[harrism]

Algorithm looks fine, just adjustments to match rest of cuDF style.

[davidwendt]

The strings API is being changed as part of the rewrite for cudf::column support. First PR is #2811.
Changes like this to the existing working code are following the APIs there for consistency.

Introducing broad cudf style changes into deprecating code may be a waste of time in my opinion. Regardless, I agree with most of these suggestions to make the API clearer until deprecation occurs.

[harrism]

If this code is being deprecated, why have the PR at all?

[davidwendt]

If this code is being deprecated, why have the PR at all?

Because it will not be replaced probably until next release and we need to keep adding features and fixing defects to the existing working functions until the new code is in place.

[harrism]

I don't think anything I suggested is onerous. If Ayush wants to skip the naming changes that's fine, but all the API, documentation, CUDF_EXPECTS etc. changes are relevant and not tedious.

[VibhuJawa]

Minor comments, everything else looks good to me.

[kkraus14]

Python looks good