Timestamp when plotting hourly time series with base != 0

[bmu]

If I plot an hourly time series with a different base (don't know if this is correct englisch I mean if it is not 12:00, 13:00, ... but something like 12:30, 13:30, ...) the values are plotted as full hours at the xaxis. If I convert to a resolution of '60min' it works.

In [250]: idx = pd.date_range('2012-12-20', periods=24, freq='H') + datetime.timedelta(minutes=30)

In [251]: idx
Out[251]: 
<class 'pandas.tseries.index.DatetimeIndex'>
[2012-12-20 00:30:00, ..., 2012-12-20 23:30:00]
Length: 24, Freq: H, Timezone: None

In [252]: df = pd.DataFrame(np.arange(24), index=idx)

In [253]: df.plot(style='r.')
Out[253]: <matplotlib.axes.AxesSubplot at 0xd34920d0>

In [254]: df2 = df.resample('60min', base=30, closed='right')

In [255]: df2.index
Out[255]: 
<class 'pandas.tseries.index.DatetimeIndex'>
[2012-12-20 00:30:00, ..., 2012-12-20 23:30:00]
Length: 24, Freq: 60T, Timezone: None

In [256]: df2.plot(style='r.')
Out[256]: <matplotlib.axes.AxesSubplot at 0xd43fde90>

[changhiskhan]

This is happening because of the internal conversion to PeriodIndex for better tick labels. We'll put in a fix for 0.10.1. Thanks.

[changhiskhan]

I'm putting in a hack for this now so for 'H' it'll keep it as DatetimeIndex and not perform the PeriodIndex conversion. DatetimeIndex has much more functionality not in the PeriodIndex so more and more restrictions are coming up in plotting in order to retain the tick labeling. Probably the answer is to redo the tick labelling so it doesn't rely on Periods at all.