BUG: series replace should allow a single list

[cpcloud]

Also reverts the method parameter of NDFrame.replace() deprecation

closes #4743

[jreback]

I feel this exists somewhere already in common?

[cpcloud]

hmm. i can't find it after grining for method and backfill, i don't see anything similar...this was in the 0.12 release but not sure when it was removed...should i move to common.py?

[jreback]

i would put it in common ; i will look for this too

[cpcloud]

@jreback this ok 2 merge? or did u find a _get_fill_func-type function elsewhere?

[jreback]

I don't think this will work if you pass odd dtypes eg 'a' to an int series or something ; maybe put some tests that change dtype on purpose

[jreback]

fyi...may want to move some of your tests to test_generic.py

[cpcloud]

is it ok if i move in a separate pr?

[cpcloud]

re strange dtypes: what behavior do u want? if you pass a replacement for a value that doesn't exist, then the original series is returned. e.g.,

In [16]: s = Series(arange(5.))

In [17]: s.replace({'a': 1, 'b': 1})
Out[17]:
0    0
1    1
2    2
3    3
4    4
dtype: float64

In [18]: s.replace(['a','b'])
Out[18]:
0    0
1    1
2    2
3    3
4    4
dtype: float64

but mixing things works

In [19]: s.replace(['a',1])
Out[19]:
0    0
1    0
2    2
3    3
4    4
dtype: float64

[jreback]

sep PR for tests ok....

what I mean is, just some confirming tests....

In [1]: s = Series(np.arange(5))

In [2]: s.replace?

In [3]: s.replace([3],[3.])
Out[3]: 
0    0
1    1
2    2
3    3
4    4
dtype: int64

In [4]: s.replace([3],[3.5])
Out[4]: 
0    0.0
1    1.0
2    2.0
3    3.5
4    4.0
dtype: float64

In [5]: s.replace([3,4],[3.5,'a'])
Out[5]: 
0      0
1      1
2      2
3    3.5
4      a
dtype: object

In [6]: s.replace([3,4],[3.5,Timestamp('20130101')])
Out[6]: 
0                      0
1                      1
2                      2
3                    3.5
4    2013-01-01 00:00:00
dtype: object

In [7]: s.replace([3,4],[3.5,True])
Out[7]: 
0    0.0
1    1.0
2    2.0
3    3.5
4    1.0
dtype: float64

note the last one; I am not sure we can do much about that (well we can but only if you use Block.setitem, where I attempt to see if you are using a bool and do the right thing)....but too much work really

[cpcloud]

ah ok

[cpcloud]

@jreback this ok?

[jreback]

I guess I don't understand how this works

ser = Series([0, 1, 2, 3, 4])
result = ser.replace([1,2,3])
assert_series_equal(result, Series([0,0,0,0,4]))

so the 1,2,3 are all replaced by 0? (shouldn't it be nan)?

[cpcloud]

It's treating 1, 2, and 3 as "missing" and replacing them by the nearest valid value (this is what the pad_1d function does)

[jreback]

ic....so tantamount to s[[1,2,3]] = np.nan; s.ffill()

[jreback]

ok then

[cpcloud]

Yes. In fact, that's what I was doing, but then dtype is not preserved (in cases where it can be)

[jreback]

cool...go ahead then

[jreback]

in test_series.py

       def test_replace_mixed_types(self):
(Pdb) l
4697            s = Series(np.arange(5))
4698 

This fails on 32-bit because s is int32 (and expected is int64)
use s = Series(np.arange(5),dtype='int64') here.....

as the numpy arrays pass thru the dtype :)

I am just going to fix it in a PR I am doing. FYI

[cpcloud]

ah ok ... cool thanks