ENH: weekday of year, common behavior accessor

[trailsquirrel]

See below. Dates from the last week of December 2013 are interpreted as being from the first week of the year. Is this due to the fact December 30th was a Monday?

import pandas as pd

dt = '2013-12-30'
pd.to_datetime(dt, coerce=True).year, pd.to_datetime(dt, coerce=True).weekofyear
# I expected: (2013, 53) since this is day 364 and the start of the 53rd week
# Instead it returned: (2013, 1)

My current work-around is:

import pandas as pd
import numpy as np

dt = '2013-12-30'
doy = pd.to_datetime(dt, coerce=True).dayofyear
(np.floor(doy / 7) + 1).astype('int')
# Returns 53

pd.show_versions()

INSTALLED VERSIONS
------------------
commit: None
python: 2.7.5.final.0
python-bits: 64
OS: Linux
OS-release: 3.11.0-15-generic
machine: x86_64
processor: x86_64
byteorder: little
LC_ALL: None
LANG: en_US.UTF-8

pandas: 0.13.1
Cython: 0.20.1
numpy: 1.8.1
scipy: 0.12.0
statsmodels: 0.4.2
IPython: 2.0.0
sphinx: 1.2.2
patsy: None
scikits.timeseries: None
dateutil: 1.5
pytz: 2012c
bottleneck: None
tables: 2.4.0
numexpr: 2.0.1
matplotlib: 1.2.1
openpyxl: 1.5.8
xlrd: 0.9.2
xlwt: 0.7.4
xlsxwriter: None
sqlalchemy: None
lxml: 3.2.0
bs4: 4.2.0
html5lib: 0.95-dev
bq: None
apiclient: None

[jreback]

related bug (not excactly dupe) of #6538

want to have a go at fixing?

[trailsquirrel]

Do you know the reason for the code that exists under the 'woy' case? It seems overly complicated, but I may be missing a corner case where my simple solution doesn't work. Don't mind helping out, but want to make sure I understand the issue.

https://github.com/pydata/pandas/blob/master/pandas/tslib.pyx#L2255

[jreback]

their are some tests in pandas/tseries/tests/test_timeseries.py

I am not sure of the reason for the estimate then verify (though maybe its the leap year issue).

you can add in your posted case as a test, have it fail, then fix (and don't break other tests).

[trailsquirrel]

So I fixed it and it broke the existing weekofyear test. After digging in, the existing code does conform to the ISO standard. Though it is not intuitive and leads to unexpected results when you try to do group bys using it.

Thoughts on adding a new DatetimeIndex attribute to help solve this? It may make sense to create an iso_weekofyear attribute from the existing to clearly distinguish them.

[jreback]

that makes sense

changing the name would require a deprecation cycle. maybe add as weekofyear_common?

open to adding

[vitorcurtis]

Sorry about the question, but where I can find weekofyear_common?

[jreback]

see #7028, but was closed as stale (not finished), you are welcome to revive.

[jorisvandenbossche]

@vitorcurtis This is an open issue, which means it is not yet implemented (update, yes see @jreback answer)

[jorisvandenbossche]

BTW, when someone wants to implement this, important question that first should be answered is what would exactly be the definition of such a 'common week'.
ISO weeks are well defined, so we need an alternative definition, see also comment on the closed PR #7028 (comment)

[vitorcurtis]

I believe that more definitions are never a good path, so my opinion is to use the definition PR #7028.
Actually, we may be flexible.
We could alert of this ISO issue and also solve the problem by simply using one or parameters for the weekofyear, as:
weekofyear(iso=False, first_weekday='Monday')
weekofyear(iso=False, first_weekday='Thursday')
...
What do you think?

[SHxKM]

Any progress with this?

[jreback]

Any progress with this?

you are welcome to submit a PR