Fix pipeline caching for older versions of joblib

[chkoar]

Reference Issue

Fixes #685

[codecov]

Codecov Report

Merging #687 into master will increase coverage by 0.08%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #687      +/-   ##
==========================================
+ Coverage   96.37%   96.45%   +0.08%     
==========================================
  Files          82       82              
  Lines        4989     4999      +10     
==========================================
+ Hits         4808     4822      +14     
+ Misses        181      177       -4     

Impacted Files	Coverage Δ
imblearn/utils/_show_versions.py	100.00% <ø> (ø)
imblearn/pipeline.py	97.91% <100.00%> (+3.97%)	⬆️
imblearn/tests/test_pipeline.py	97.68% <100.00%> (+0.03%)	⬆️
imblearn/utils/tests/test_show_versions.py	100.00% <100.00%> (ø)
imblearn/ensemble/tests/test_forest.py	100.00% <0.00%> (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 703cee1...fcac0d9. Read the comment docs.

[pep8speaks]

Hello @chkoar! Thanks for updating this PR. We checked the lines you've touched for PEP 8 issues, and found:

There are currently no PEP 8 issues detected in this Pull Request. Cheers! 🍻

Comment last updated at 2020-06-08 18:30:19 UTC

[glemaitre]

Why this change?

[chkoar]

As a test be cause I was getting errors about FutureWarnings. I believe those tests should run only on cron based tasks because otherwise we get unrelated errors to PRs. What to you think?

[glemaitre]

We are running those in scikit-learn. It allows to catch the warnings and ensure that you don't introduce new one by calling deprecated stuff. If this is in test, you can always filter a specific warning.

[chkoar]

So, what do you propose? I think that the main thing is not the deprecation and future warnings as errors. We should run those in periodic manner against the latest version of scikit-learn otherwise we are getting errors that are irrelevant to PRs. So, we could test against the latest (released) version of scikit-learn and even a previous one. The latest could be tested in schedule. Or something like that.

[chkoar]

This PR could be merged after #691