VIS/BUG DatetimeIndex with freq=None plot return rotated xtickslabels

[weber-s]

Code Sample

file = 'test.csv' #with a "date" column
df1  = pd.read_csv(file, index_col="date", parse_dates=["date"])
df1.plot()

idx =pd.date_range('2000-01-01','2003-01-01', freq="6d")
df2 = pd.DataFrame(index=idx, data=np.random.randn(len(idx),1))
df2.plot()

Problem description

When plotting a time indexed series, we expect no rotation in x labels. It does not work when importing data with

df  = pd.read_csv(file, index_col="date", parse_dates=["date"])

where the rotation occurs. Moreover, set rot=0 gives a strange behavior with label not centered on ticks.

The only difference is that freq=None in the first case.

Expected Output

No rotation by default.

Here is what we have for df1.plot() (Problem) :

Here is what we have for df1.plot(rot=0) (Problem) :

Here is what we have for df2.plot() (Ok) :

Details

May be related to :

• VIS: datetime plots x axis format: labels now rotated? #8150
• VIS/BUG: subplots=True gives rotated tick labels #15025

INSTALLED VERSIONS ------------------ commit: None python: 3.6.0.final.0 python-bits: 64 OS: Linux OS-release: 4.9.9-1-ARCH machine: x86_64 processor: byteorder: little LC_ALL: None LANG: fr_FR.UTF-8 LOCALE: fr_FR.UTF-8

pandas: 0.19.2
nose: None
pip: None
setuptools: 34.2.0
Cython: None
numpy: 1.12.0
scipy: 0.18.1
statsmodels: None
xarray: None
IPython: 5.1.0
sphinx: None
patsy: None
dateutil: 2.6.0
pytz: 2016.10
blosc: None
bottleneck: None
tables: None
numexpr: None
matplotlib: 2.0.0
openpyxl: None
xlrd: None
xlwt: None
xlsxwriter: None
lxml: None
bs4: None
html5lib: None
httplib2: 0.10.3
apiclient: None
sqlalchemy: None
pymysql: None
psycopg2: None
jinja2: 2.9.5
boto: None
pandas_datareader: None

[jorisvandenbossche]

The rotation is done on purpose, as you would otherwise get overlapping labels (like in your second example plot, only without the non-centered labels).

So there are two ways pandas plots datetimes under the hood: ones with a regular frequence, and the ones without regular frequency (that is the reason you see a difference for frames with a different freq). The problem is that only for the regular time series, we have custom label formatting (your third plot). For the non-regular, we just use matplotlib default formatting, with this difference that we apply a rotation by default to have non-overlapping labels.
See #15071 for a longer explanation.

The shift in the labels (non-centered) when using rot=0 is a bit annoying.

[weber-s]

Thanks for the explanation (personally, I am in favour of passing all plots in period base plotting, it looks nicer).

The shift in the labels (non-centered) when using rot=0 is a bit annoying.

I spoted the problem in format_date_labels:

def format_date_labels(ax, rot):
    # mini version of autofmt_xdate
    try:
        for label in ax.get_xticklabels():
            label.set_ha('right') # here is the problem
            label.set_rotation(rot)
        fig = ax.get_figure()
        fig.subplots_adjust(bottom=0.2)
    except Exception:  # pragma: no cover
        pass

with rot=0 the alignement should be center. However I have no idea how to solve it.

I think it's a feature (having the end of the labels on the ticks), however it is a bit confusing without rotation.

[mroeschke]

Seems like this behavior is intended. Going to close for now, but happy to reopen if there is future interest on an enhancement on this behavior.