fix(steps): add unit test and fix unique col scaling

[jitingxu1]

changes:

1. fix unique col standardization
2. add unit test

resolve #119

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 89.73%. Comparing base (6dce35e) to head (b49543b).

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #158      +/-   ##
==========================================
+ Coverage   88.70%   89.73%   +1.03%     
==========================================
  Files          27       28       +1     
  Lines        2009     2045      +36     
==========================================
+ Hits         1782     1835      +53     
+ Misses        227      210      -17     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[deepyaman]

This doesn't quite fix it; it throws an error. I'm not too keen on it.

Why? Preprocessing can be like a recipe you apply, and you may always want to scale your numeric columns. I don't want to throw an error just because one of those columns is not unique; should figure out how to handle it instead.

See how scikit-learn handles it, for example: https://github.com/betatim/scikit-learn/blob/main/sklearn/preprocessing/_data.py#L82 (huh, not sure how I got to somebody's fork, but whatever... get the point)

If you want to refactor this PR to just add the unit tests, and create an issue to handle scaling without throwing an error, we can do that (but doesn't need to go in today).

(Or you can push off adding the unit tests until address the scaling in a future PR)

[jitingxu1]

This doesn't quite fix it; it throws an error. I'm not too keen on it.

Why? Preprocessing can be like a recipe you apply, and you may always want to scale your numeric columns. I don't want to throw an error just because one of those columns is not unique; should figure out how to handle it instead.

See how scikit-learn handles it, for example: https://github.com/betatim/scikit-learn/blob/main/sklearn/preprocessing/_data.py#L82 (huh, not sure how I got to somebody's fork, but whatever... get the point)

If you want to refactor this PR to just add the unit tests, and create an issue to handle scaling without throwing an error, we can do that (but doesn't need to go in today).

(Or you can push off adding the unit tests until address the scaling in a future PR)

make sense, I will update this PR soon.

[jitingxu1]

numpy could get the machine precision for different float precision. I took the float16 precision as all the approximate precision for all float types to bypass the use of numpy.