PERF: sort_values speedup for multiple columns with random numeric values

[mficek]

Continue work on #17141.
Speedup achieved for sorting by multiple columns: lexsort over multiple columns
replaced with mergesort of columns in reverse order than specified in "by"
argument of DataFrame.sort_values().
Issues:

• only works for numerical data

• only random enough data achieve sorting speedup (no factorization)

• ASV of one-column sort is worse, why?

• closes #xxxx

• tests added / passed

• passes git diff upstream/master -u -- "*.py" | flake8 --diff

• whatsnew entry

[pep8speaks]

Hello @mficek! Thanks for updating the PR.

Cheers ! There are no PEP8 issues in this Pull Request. 🍻

Comment last updated on January 14, 2018 at 08:37 Hours UTC

[mficek]

I'm running tests and found out, that many of them fail simply because expected method for multi column sort is not a mergesort (stable) which I use. Hm... Should I start fixing tests? Doesn't sound right to me, but the approach I use is imho correct.

[jreback]

I'm running tests and found out, that many of them fail simply because expected method for multi column sort is not a mergesort (stable) which I use. Hm... Should I start fixing tests? Doesn't sound right to me, but the approach I use is imho correct.

can you show what you mean here?

prob would just change the default (this behavior is inherited from ndarray which defaults to quicksort :<)

[desasaliho]

Hello..i have a problem in my Facebook when I login it say your account has been disabled
By an administrator I want to know why I am Disabled please and when I can login back I
Hope that you can help me and it thank you for
Any answer you will give it to me bye

[jreback]

can you rebase and fixup

[mficek]

Hi Jeff, I'm sorry, but I can not find enough time and motivation to finish this issue. It seems more complex than expected with improvements being too much dependent of the nature of the data.