ENH dataframe in/out for all samplers

[glemaitre]

Reference Issue

closes #639

What does this implement/fix? Explain your changes.

Let dataframe out if they got in.

TODO:

• Add some documentation.

Any other comments?

[lgtm-com]

This pull request introduces 1 alert when merging e936bdd into 45b538c - view on LGTM.com

new alerts:

• 1 for Mismatch in multiple assignment

[codecov]

Codecov Report

Merging #644 into master will increase coverage by 0.54%.
The diff coverage is 100%.

@@            Coverage Diff            @@
##           master    #644      +/-   ##
=========================================
+ Coverage   97.96%   98.5%   +0.54%     
=========================================
  Files          83      82       -1     
  Lines        4867    4888      +21     
=========================================
+ Hits         4768    4815      +47     
+ Misses         99      73      -26

Impacted Files	Coverage Δ
imblearn/over_sampling/_smote.py	97.23% <100%> (ø)	⬆️
imblearn/utils/estimator_checks.py	96.11% <100%> (+7.22%)	⬆️
imblearn/over_sampling/tests/test_smote_nc.py	100% <100%> (ø)	⬆️
...ling/_prototype_selection/_random_under_sampler.py	100% <100%> (ø)	⬆️
imblearn/over_sampling/_random_over_sampler.py	100% <100%> (ø)	⬆️
...rototype_selection/_instance_hardness_threshold.py	100% <0%> (ø)	⬆️
...election/tests/test_instance_hardness_threshold.py	100% <0%> (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 45b538c...2f93b40. Read the comment docs.

[chkoar]

I thought that you were against on using pandas. With your implementation you assume that the user has pandas installed. Of course it is passing pandas pandas df. Shouldn't we try/catch in case that pandas in not installed and is passing something that has a loc property?

[glemaitre]

I am still against. Before sampling I am duck typing. At the moment to recreate the database if we have the columns name it means that we gave us a database meaning that the user has pandas. So in this case only I am importing pandas to call the dataframe constructor. Pandas can still be an optional dependence then.

[lgtm-com]

This pull request introduces 1 alert when merging b1d2d56 into 45b538c - view on LGTM.com

new alerts:

• 1 for Mismatch in multiple assignment

[lgtm-com]

This pull request introduces 1 alert when merging 9431bb8 into 45b538c - view on LGTM.com

new alerts:

• 1 for Mismatch in multiple assignment

[glemaitre]

@chkoar Would you have a bit of time too look at it.

[lgtm-com]

This pull request introduces 1 alert when merging 2f93b40 into 153f6e0 - view on LGTM.com

new alerts:

• 1 for Mismatch in multiple assignment

[chkoar]

I will give it a go soon

[chkoar]

Why not doing this in _check_X_y?

[glemaitre]

This block required the resampled data so _check_X_y is called before that the resampling happened.

[chkoar]

Ok. It is like we need to extracted it in an internal transformation method. But I am ok as is.

[chkoar]

Is there a case where something that will pass, it will have .columns, .loc and will not be a DataFrame? Probably is a corner case. In that case we will have an ugly error. Is there a need to test/catch this? If no then I think I am ok.

[glemaitre]

Is there a case where something that will pass, it will have .columns, .loc and will not be a DataFrame? Probably is a corner case. In that case we will have an ugly error. Is there a need to test/catch this? If no then I think I am ok.

We always ducktype like this in scikit-learn for dataframe. So if it passes in imblearn it will break in scikit-learn just with the _safe_indexing.

[chkoar]

Go ahead!