Python, Pandas ; ValueError('window must be an integer',)

Question

I seem to be having this issue with Pandas code inside a Bokeh callback.

Here's part of the output before the error. My dataframe seems normal and I'm not sure why it's upset

                     time  temperature
0 2016-03-17 11:00:00        4.676
1 2016-03-17 11:30:00        4.633
2 2016-03-17 12:00:00        4.639
3 2016-03-17 12:30:00        4.603
4 2016-03-17 13:00:00        4.615
5 2016-03-17 13:30:00        4.650
6 2016-03-17 14:00:00        4.678
7 2016-03-17 14:30:00        4.698
8 2016-03-17 15:00:00        4.753
9 2016-03-17 15:30:00        4.847
ERROR:bokeh.server.protocol_handler:error handling message Message 'PATCH-DOC' (
revision 1): ValueError('window must be an integer',)

And here's the code I changed from the flask embed example (link here):

def callback(attr, old, new):
        df = pd.DataFrame.from_dict(source.data.copy())
        print df[:10]
        if new == 0:
            data = df
        else:
            data = df.rolling('{0}D'.format(new)).mean()
        source.data = ColumnDataSource(data=data).data

    slider = Slider(start=0, end=30, value=0, step=1, title="Smoothing by N Days")
    slider.on_change('value', callback)

I can also include the full code if that help, but the main change I have is just a doc.add_periodic_callback() that fetches new data periodically.

Answer 1

This is an error from Pandas. You are passing a string to df.rolling, but it expects only integer values. You probably want to pass int(new) instead.

Edit: as noted below, evidently the Pandas documentation is incomplete, and the real ultimate problem in this case is probably the lack of a time index, since creating a naive Dataframe and passing values like "10d" definitely raises the indicated error:

In [2]: df = pd.DataFrame({'B': [0, 1, 2, 10, 4]})

In [3]: df.rolling('10d')
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-3-2a9875316cd7> in <module>
----> 1 df.rolling('10d')

~/anaconda/lib/python3.7/site-packages/pandas/core/generic.py in rolling(self, window, min_periods, center, win_type, on, axis, closed)
   8906                                    min_periods=min_periods,
   8907                                    center=center, win_type=win_type,
-> 8908                                    on=on, axis=axis, closed=closed)
   8909
   8910         cls.rolling = rolling

~/anaconda/lib/python3.7/site-packages/pandas/core/window.py in rolling(obj, win_type, **kwds)
   2467         return Window(obj, win_type=win_type, **kwds)
   2468
-> 2469     return Rolling(obj, **kwds)
   2470
   2471

~/anaconda/lib/python3.7/site-packages/pandas/core/window.py in __init__(self, obj, window, min_periods, center, win_type, axis, on, closed, **kwargs)
     78         self.win_freq = None
     79         self.axis = obj._get_axis_number(axis) if axis is not None else None
---> 80         self.validate()
     81
     82     @property

~/anaconda/lib/python3.7/site-packages/pandas/core/window.py in validate(self)
   1476
   1477         elif not is_integer(self.window):
-> 1478             raise ValueError("window must be an integer")
   1479         elif self.window < 0:
   1480             raise ValueError("window must be non-negative")

ValueError: window must be an integer

Answer 2

df.rolling can also handle time periods. Make sure the date time is in pandas format. If not, convert as such -

data['col'] = pd.to_datetime(data['col'])

Answer 3

As of today, the documentation states as follows:

window : int, or offset

Size of the moving window. This is the number of observations used for calculating the statistic. Each window will be a fixed size.

If its an offset then this will be the time period of each window. Each window will be a variable sized based on the observations included in the time-period. This is only valid for datetimelike indexes. This is new in 0.19.0

It is not clear for me whether the time information in your dataframe is a column or part of a MultiIndex. For the first case, you can use .set_index('time').

For MultiIndex, currently, you cannot use offsets. See the related issue. Instead, you can use .reset_index() to transform it into a single index dataframe (see here).

Update: you can also pass datetime columns for offset-based rolling metrics with the on parameter (and, therefore, you do not have to have them in the index).