Using "index" in a Series with a dict produces results inconsistent with documentation.

[jsal13]

Code Sample, a copy-pastable example if possible

# The first case shows that dicts cause the index argument to be more of a WHERE selector for rows...

>>> pd.Series({"a": 1, "b": 2, "c": 3}, index=["a", "b"])
a    1
b    2
dtype: int64

# But the docs + using a list have an error thrown if the index size does not match the length of the data.
>>> pd.Series(["a", "b", "c"], index=[0, 1])
# Error...
ValueError: Length of passed values is 3, index implies 2

Problem description

The current behavior with dicts is inconsistent with the documentation. Moreover, there should be consistent behavior between passing a dict and passing an array into the series.

Expected Output

Output of pd.show_versions()

[paste the output of pd.show_versions() here below this line]
INSTALLED VERSIONS

commit : None
python : 3.7.3.final.0
python-bits : 64
OS : Darwin
OS-release : 17.7.0
machine : x86_64
processor : i386
byteorder : little
LC_ALL : None
LANG : en_US.UTF-8
LOCALE : en_US.UTF-8

pandas : 0.25.1
numpy : 1.17.2
pytz : 2019.2
dateutil : 2.8.0
pip : 19.1.1
setuptools : 41.2.0
Cython : 0.29.12
pytest : 5.0.1
hypothesis : None
sphinx : 2.1.2
blosc : None
feather : None
xlsxwriter : 1.1.8
lxml.etree : 4.3.4
html5lib : 1.0.1
pymysql : None
psycopg2 : None
jinja2 : 2.10.1
IPython : 7.6.1
pandas_datareader: None
bs4 : 4.7.1
bottleneck : 1.2.1
fastparquet : None
gcsfs : None
lxml.etree : 4.3.4
matplotlib : 3.1.0
numexpr : 2.6.9
odfpy : None
openpyxl : 2.6.2
pandas_gbq : None
pyarrow : None
pytables : None
s3fs : None
scipy : 1.3.0
sqlalchemy : 1.3.8
tables : 3.5.2
xarray : None
xlrd : 1.2.0
xlwt : 1.3.0
xlsxwriter : 1.1.8

[TomAugspurger]

The current behavior with dicts is inconsistent with the documentation

Can you point to the docs that you think are inconsistent?

I think the dict-case is essentially reindex, matching the behavior of Series(series, index).

[jsal13]

https://pandas.pydata.org/pandas-docs/stable/reference/api/pandas.Series.html

index : array-like or Index (1d)
Values must be hashable and have the same length as data. Non-unique index values are allowed. Will default to RangeIndex (0, 1, 2, …, n) if not provided. If both a dict and index sequence are used, the index will override the keys found in the dict.

The first bolded statement is incorrect currently since (for example) the first example above shows that you can have an index strictly shorter than the length of the data. The second statement is not correct either, as I interpret it: using this should "override the keys found in the dict" --- I'm not exactly sure what this is supposed to mean, but what it does is selects elements that match that index key.

>>> pd.Series({"a": 1, "b": 2, "c": 3}, index=["x", "y", "z"])
x   NaN
y   NaN
z   NaN
dtype: float64

I would expect this, given the docs, to rename "a" to "x" and so forth.

---

The dict case is reindex, but it does not match the behavior of Series(array, index) when len(index) < len(array) --- in particular, it does not throw an error. In fact, even when the len(index) == len(array), we have that they behave differently, as the above example shows, since changing the dict to an array [1, 2, 3] will give a series 1, 2, 3 with an index of "x", "y", "z" respectively.

[TomAugspurger]

Values must be hashable and have the same length as data

OK, thanks. That is incorrect. We do allow different lengths when the data has labels (Series or dict-like).

I would expect this, given the docs, to rename "a" to "x" and so forth.

Gotcha. Your expectation is incorrect. By "override" we mean reindex.

[jsal13]

Got'cha. Okay, so this may not be an issue anymore if the docs change. Is this intended behavior (in general), though? It feels significantly different depending on if you use an array vs. a dict, each with the index parameter.

[TomAugspurger]

Yes, it's intended. Notice that dict and list differ when not using `index`. The keys of the dict are used to form the index. We don't want to just ignore those when `index` is passed. I think that a PR correcting the statement that the lengths of `index` and `data` must match would be most welcome.

…

On Thu, Sep 12, 2019 at 2:03 PM James ***@***.***> wrote: Got'cha. Okay, so this may not be an issue anymore if the docs change. Is this intended behavior (in general), though? It feels significantly different depending on if you use an array vs. a dict, each with the index parameter. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#28418?email_source=notifications&email_token=AAKAOISPKU4FUWOHUAQNBGDQJKHAZA5CNFSM4IWHBZS2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD6S5NKI#issuecomment-530962089>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAKAOIV7UOKN2UFBZ7ZJJZ3QJKHAZANCNFSM4IWHBZSQ> .

[jsal13]

Alright, I'll make a PR for that and close my other PR. Thanks!