Implements Series.filter

[beobest2]

Implementing Series.filter

>>> kser = ks.Series([0, 1, 2], index=['one', 'two', 'three'])

>>> kser
one      0
two      1
three    2
Name: 0, dtype: int64

>>> kser.filter(items=['one', 'three'])
one      0
three    2
Name: 0, dtype: int64

>>> kser.filter(regex='e$')
one      0
three    2
Name: 0, dtype: int64

>>> kser.filter(like='hre')
three    2
Name: 0, dtype: int64

[codecov-commenter]

Codecov Report

Merging #1511 into master will increase coverage by 0.00%.
The diff coverage is 98.66%.

@@           Coverage Diff           @@
##           master    #1511   +/-   ##
=======================================
  Coverage   94.19%   94.20%           
=======================================
  Files          38       38           
  Lines        8595     8608   +13     
=======================================
+ Hits         8096     8109   +13     
  Misses        499      499           

Impacted Files	Coverage Δ
databricks/koalas/frame.py	95.68% <ø> (-0.08%)	⬇️
databricks/koalas/missing/series.py	100.00% <ø> (ø)
databricks/koalas/generic.py	97.05% <98.66%> (+0.45%)	⬆️

---

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 3f08d80...b3791e3. Read the comment docs.

[HyukjinKwon]

Ah, #1512 caused a conflict here. Can you rebase and resolve the conflicts please?

[beobest2]

@HyukjinKwon please check this.

1. Resolving code conflicts
2. Remove filter of frame.py
3. Add multi-index support implemented at 4a18a09
4. Add multi-index test cases in Series

[HyukjinKwon]

LGTM, thanks @beobest2

[ueshin]

Sorry for the late review.
Actually I'd say it's not good to move the function to generic.
I'll submit a PR to fix it soon.

Btw, this broke the docstring, I'll submit the PR to fix it anyway.

[HyukjinKwon]

Ah .. I didn't notice the newlines disappeared here.. it breaks the doc rendering. Let's be careful next time.

[beobest2]

Oppse I'll be careful next time. I'm Sorry

[HyukjinKwon]

Okay, I had a offline discussion. We have been following pandas' structure so far. However, now it became kind of a bit weird for some instances in generic.py as we should handle both differently with an if-else.
We can avoid an if-else on self in generic.py which actually looks a bit odds, but only share the common codes.

These will be cleaned up in a separate PR.

[itholic]

@HyukjinKwon Thanks for sharing, makes sense.

[ueshin]

I don't look into all of the function in generic.py file yet so I'm not sure we should move all of them to the subclasses, but here is a pattern especially I'd avoid in generic.py:

def some_function(self, ...):

    ... preprocess ...

    if isinstance(self, Series):
        kdf = self.to_frame()
    else:
        kdf = self

    ... working with kdf and the result is ret ...

    if isinstance(self, Series):
        return first_series(ret)
    else:
        return ret

Then we should move it to DataFrame and make it work with only DataFrame, then we can add something like the following in Series:

def some_function(self, ...):

    ... preprocess ...

    return first_series(self.to_frame().some_function(...))

[ueshin]

Anyway I'll submit PRs when I encounter the pattern.