BUG: np.round does not work on a series in 0.18.0

[wholmgren]

First, conda install pandas is currently pulling down 0.18.0, which was unexpected since I only see RCs right now, but, on to the real issue...

In 0.18.0 I cannot pass a series to np.round...

In [35]: np.round(pd.Series([1.11, 2.21]), 1)
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-35-4cc66a785b99> in <module>()
----> 1 np.round(pd.Series([1.11, 2.21]), 1)

/Users/holmgren/miniconda3/envs/pvlib35/lib/python3.5/site-packages/numpy/core/fromnumeric.py in round_(a, decimals, out)
   2784     except AttributeError:
   2785         return _wrapit(a, 'round', decimals, out)
-> 2786     return round(decimals, out)
   2787
   2788

TypeError: round() takes from 1 to 2 positional arguments but 3 were given

np.round does still work on a DataFrame in 0.18.0...

In [5]: np.round(pd.DataFrame([1.11,2.22]), 1)
Out[5]:
     0
0  1.1
1  2.2

A Series on pandas 0.17.1...

In [4]: np.round(pd.Series([1.12, 2.22]), 1)
Out[4]:
0    1.1
1    2.2
dtype: float64

Ok, I know that there's now a round method in pandas (which is great), but I like to use np.round in unit tests so that I can test against earlier versions of pandas.

Version info:

In [34]: pd.show_versions()

INSTALLED VERSIONS
------------------
commit: None
python: 3.5.1.final.0
python-bits: 64
OS: Darwin
OS-release: 15.3.0
machine: x86_64
processor: i386
byteorder: little
LC_ALL: None
LANG: en_US.UTF-8

pandas: 0.18.0
nose: 1.3.7
pip: 8.1.0
setuptools: 20.2.2
Cython: None
numpy: 1.10.4
scipy: 0.17.0
statsmodels: None
xarray: None
IPython: 4.1.1
sphinx: 1.3.5
patsy: None
dateutil: 2.4.2
pytz: 2015.7
blosc: None
bottleneck: None
tables: None
numexpr: None
matplotlib: 1.5.1
openpyxl: None
xlrd: None
xlwt: None
xlsxwriter: None
lxml: None
bs4: None
html5lib: None
httplib2: None
apiclient: None
sqlalchemy: None
pymysql: None
psycopg2: None
jinja2: 2.8
boto: None

[jreback]

this is the same issue as in #12238 and solved for np.searchsorted in #12413 by @gfyoung

The issue is numpy insists on call with positional arguments a 'sub-class' method, even though we are duck like, to implement things like this, which is fine. but its just not very friendlty.

so we need to accept (but not advertise the out) arg (which was taken out in 0.18.0).

separately, @gfyoung I think numpy should have a try: except: around _wrap_it to guard against things like this.

[jreback]

0.18.0 has been released to conda, and I just put the docs up, but hasn't gone to PyPi yet, nor release email; but thanks for the report.

[gfyoung]

1. I'll grab this issue and fix it. Shouldn't be too difficult to fix on the pandas end.
2. For the try-except block, what would you suggest for the except part?

[jreback]

@gfyoung same thing as if .round doesn't exist. The problem is numpy is trying to dispatch to something which is says, hey you have a .round method AND you must use this signature (which numpy uses), which is pretty incompatible with anything remotely duck-like.

[gfyoung]

Oh, oh, okay, I see what you mean. I have a PR opened against numpy that updates the function signatures, but I'll update the PR to make sure the behaviour is consistent across all of those fromnumeric.py functions to implement that try-except functionality you are describing (I think some of them do but not all of them AFAIK)

[gfyoung]

Also, for future reference, what's the oldest numpy version that this 0.18.0 release will support?

[wholmgren]

thanks @jreback and @gfyoung!

[jreback]

@gfyoung numpy 1.7

[Dr-Irv]

I removed the "out" argument when implementing support for built-in round based on discussion here:, where @jreback said: "so drop the out paramater from Series.round (you can put that in API changes). We don't support this ANYWHERE else so its really inconsitent."

[gfyoung]

@Dr-Irv : Yeah, unfortunately numpy didn't get the memo. 😄 However, it's a problem that's a part of a much bigger issue with compatibility between these two libraries.

[jreback]

@Dr-Irv no worries, it is the right thing to do. as @gfyoung its a bigger issue which has recently surfaced in laster versions of numpy.

[Twizzledrizzle]

Are we having problem with this bug as well? On my computer with the latest Anaconda, win7 it works fine. But on a collegue's, we started getting error numpy round() takes at most 2 arguments (3 given) when he installed the latest version.

this for example works for me, but not for him

import pandas as pd
import numpy as np
a = [1.23, 1.33]
print np.round(pd.Series(a), 1)

As a work-around, making below change solved it, but I do not know if I have broken something else. Our scripts seems to work though

[gfyoung]

@Twizzledrizzle : Unfortunately, the fix isn't that simple. Your change is actually modifying fromnumeric.py, which is in the numpy library and not in the pandas library. Perhaps this works for you if you're never sending the result to another array out, but universally, this will not work I'm afraid because it modifies a library that we have no control over for all intensive purposes.

[Twizzledrizzle]

In what version do you think series will be compatible with np.round again? Thank you for all your time on this! Still using our ugly fix in fromnumeric.py

Or is it recommended to use series.round(...) instead? Unfortunately many places to fix in that case in our codebase.

[gfyoung]

Should be compatible in 0.18.1 but I would recommend using series round anyways (sorry! 😞), as maintaining total compatibly with numpy's implementation is not something we want to do for this library in light of discussions we have had previously.

[Twizzledrizzle]

Thank you gfyoung! I will try some find & replace magic