10

I created a plot which looks likeenter image description here

I have a few issues:

  1. How can i specifically show the weekends. Some ways i had thought were to grab the indices corresponding to weekends and then draw transparent bars between xlims. Also rectangle could be drawn for the same. It would be best if it could be done plainly in Pandas.
  2. The date formatting is not the most pretty

Following is the code used to generate this plot

ax4=df4.plot(kind='bar',stacked=True,title='Mains 1 Breakdown');
ax4.set_ylabel('Power (W)');
idx_weekend=df4.index[df4.index.dayofweek>=5]
ax.bar(idx_weekend.to_datetime(),[1800 for x in range(10)])

The ax.bar is specifically for highlighting weekends, but it does not produce any visible output. (Problem 1) For Problem 2 i tried to use Major Formatter and Locators, the code is as follows:

ax4=df4.plot(kind='bar',stacked=True,title='Mains 1 Breakdown');
ax4.set_ylabel('Power (W)');
formatter=matplotlib.dates.DateFormatter('%d-%b');
locator=matplotlib.dates.DayLocator(interval=1);
ax4.xaxis.set_major_formatter(formatter);
ax4.xaxis.set_major_locator(locator);

The output produced is as follows: enter image description here

It may be helpful to know what the Dataframe looks like

In [122]:df4

Out[122]:
<class 'pandas.core.frame.DataFrame'>
DatetimeIndex: 36 entries, 2011-04-19 00:00:00 to 2011-05-24 00:00:00
Data columns:
(0 to 6 AM) Dawn          19  non-null values
(12 to 6 PM) Dusk         19  non-null values
(6 to 12 Noon) Morning    19  non-null values
(6PM to 12 Noon) Night    20  non-null values
dtypes: float64(4)
Nipun Batra
  • 11,007
  • 11
  • 52
  • 77
  • It's not very complicated to achieve this in matplotlib if e.g. marking the ticklabels of the weekends in another color is an accepted solution. To filer out the weekends simpy use matplotlibs `WeekdayLocator`. Personally I think it's way easier to customize plots if they are plotted in matplotlib rather than pandas. – sodd May 18 '13 at 14:14
  • @nordev: Added this also to the solution below which now is in community wiki – Nipun Batra May 18 '13 at 16:46

2 Answers2

13

I tried a lot and for now these hacks work. Await a more Pythonic and consistent solutions. Solution to labeling problems:

def correct_labels(ax):
    labels = [item.get_text() for item in ax.get_xticklabels()]
    days=[label.split(" ")[0] for label in labels]
    months=["Jan","Feb","Mar","Apr","May","Jun","Jul","Aug","Sep","Oct","Nov","Dec"]
    final_labels=[]
    for i in range(len(days)):
        a=days[i].split("-")
        final_labels.append(a[2]+"\n"+months[int(a[1])-1])
    ax.set_xticklabels(final_labels)

Also while plotting i make the following change

ax=df.plot(kind='bar',rot=0)

This makes the labels at 0 rotation.

For finding weekends and highlighting them, i wrote the following two functions:

def find_weekend_indices(datetime_array):
    indices=[]
    for i in range(len(datetime_array)):
        if datetime_array[i].weekday()>=5:
            indices.append(i)
    return indices

def highlight_weekend(weekend_indices,ax):
    i=0
    while i<len(weekend_indices):
         ax.axvspan(weekend_indices[i], weekend_indices[i]+2, facecolor='green', edgecolor='none', alpha=.2)
         i+=2

Now, the plot looks much more useful and covers these use cases.enter image description here

sodd
  • 12,482
  • 3
  • 54
  • 62
Nipun Batra
  • 11,007
  • 11
  • 52
  • 77
  • 1
    I would recommend to use `ax.axvspan(weekend_indices[i], weekend_indices[i]+2, facecolor='green', edgecolor='none', alpha=.2)` to create the green "fillings", as one does not have to explicilty specify the `ymax`. – sodd May 18 '13 at 16:59
  • Works for time series objects only when replacing `indices.append(i)` by `indices.append(datetime_array[i])`. – Martin Dec 12 '18 at 22:41
6

Now that Pandas supports the powerful .dt namespace on every series, it is possible to identify the start and end of each weekend without any explicit Python loops. Simply filter your time values with t.dt.dayofweek >= 5 to select only times falling on the weekend, and then group by a made-up value that is different every week — here I use year * 100 + weekofyear because the result looks like 201603 which is fairly pleasant to read for debugging.

The resulting function is:

def highlight_weekends(ax, timeseries):
    d = timeseries.dt
    ranges = timeseries[d.dayofweek >= 5].groupby(d.year * 100 + d.weekofyear).agg(['min', 'max'])
    for i, tmin, tmax in ranges.itertuples():
        ax.axvspan(tmin, tmax, facecolor='orange', edgecolor='none', alpha=0.1)

Simply pass it the axis and the time series that is your x axis, and it will highlight the weekends for you!

Brandon Rhodes
  • 83,755
  • 16
  • 106
  • 147