CODEWITHSUNDEEP
pythonnumpypandastime-series
hernanavella
hernanavella

Reputation: 5552

How to do resample of intraday timeseries data with dateOffset in Pandas/Numpy?

I'm dealing with futures data, where the current day starts before 00:00:00. I need to do resampling of 1 minute data to 1 hour data, taking into account the date offset. Let's see an example:

df1 - 1min data

                        Open     High      Low    Close
2005-09-06 17:27:00  1234.75  1234.75  1234.75  1234.75
2005-09-06 17:28:00  1234.75  1234.75  1234.75  1234.75
2005-09-06 17:29:00  1234.75  1234.75  1234.75  1234.75
2005-09-06 17:30:00  1234.75  1234.75  1234.50  1234.50
2005-09-06 18:01:00  1234.50  1234.50  1234.25  1234.50
2005-09-06 18:02:00  1234.50  1234.50  1234.50  1234.50
2005-09-06 18:03:00  1234.50  1234.50  1234.50  1234.50
2005-09-06 18:04:00  1234.50  1234.50  1234.50  1234.50
2005-09-06 18:05:00  1234.50  1234.50  1234.25  1234.25

This is what happens with a regular resample:

conversion = {'Open': 'first', 'High': 'max', 'Low': 'min', 'Close': 'last'}

In [77]: normal_resample = df1.resample(rule='60Min', how=conversion)
In [78]: normal_resample
Out[79]: 
                        Open     High      Low    Close
2005-09-06 17:00:00  1234.75  1234.75  1234.50  1234.50
2005-09-06 18:00:00  1234.50  1234.50  1234.25  1234.25

This is the desired output accounting for the offset:

conversion = {'Open': 'first', 'High': 'max', 'Low': 'min', 'Close': 'last'}

In [77]: offset_resample = df1.resample(rule='60Min', how=conversion) + offset:18:00
In [78]: offset_resample
Out[79]: 
                        Open     High      Low    Close
2005-09-06 17:00:00  1234.75  1234.75  1234.50  1234.50
2005-09-07 18:00:00  1234.50  1234.50  1234.25  1234.25

Notice how what I want is specify that the new day starts at '18:00:00', not at midnight.

What have I done: I have a formula that includes the offset, resampling to daily data, but I don't know how to adapt it for intraday resampling.

def resample_day(df):
    df.index = pd.DatetimeIndex(df.index)
    df = df.between_time('18:00', '16:00', include_start=True, include_end=True)
    proxy = df.index + pd.DateOffset(hours=6)  # This is the piece that does the trick
    result = df.groupby(proxy.date).agg(
        {'Open': 'first', 'High': 'max', 'Low': 'min', 'Close': 'last'})
    result = result.reindex(columns=['Open', 'High', 'Low', 'Close'])
    return result

Also, what is suggested here, doesn't work, for me at least.

Thanks for the input.

Upvotes: 3

Views: 1580

Answers (1)

Andy Hayden
Andy Hayden

Reputation: 375475

You could just add on a day (if it the time is past 5pm):

In [11]: df1.index.time > datetime.time(17)
Out[11]: array([False,  True], dtype=bool)

In [12]: df1.index + np.where((df1.index.time > datetime.time(17)), pd.offsets.Day(1).nanos, 0)
Out[12]:
<class 'pandas.tseries.index.DatetimeIndex'>
[2005-09-06 17:00:00, 2005-09-07 18:00:00]
Length: 2, Freq: 60T, Timezone: None

Upvotes: 2

Related Questions