Filtering pandas dataframe by day

Question

I have a pandas data frame with forex data by minutes, one year long (371635 rows):

                           O        H        L        C
0                                                      
2017-01-02 02:00:00  1.05155  1.05197  1.05155  1.05190
2017-01-02 02:01:00  1.05209  1.05209  1.05177  1.05179
2017-01-02 02:02:00  1.05177  1.05198  1.05177  1.05178
2017-01-02 02:03:00  1.05188  1.05200  1.05188  1.05200
2017-01-02 02:04:00  1.05196  1.05204  1.05196  1.05203

I want to filter daily data to get an hour range:

dt = datetime(2017,1,1)
df_day = df1[df.index.date == dt.date()]
df_day_t = df_day.between_time('08:30', '09:30')   

If I do a for loop with 200 days, it takes minutes. I suspect that at every step this line

df_day = df1[df.index.date == dt.date()] 

is looking for the equality with every row in the data set (even if it is an ordered data set).

Is there any way I could speed up the filtering or I should just do some old imperative for loop from January to December...?

Answer 1

Avoid Python datetime

First you should avoid combining Python datetime with Pandas operations. There are many Pandas / NumPy friendly methods to create datetime objects for comparison, e.g. pd.Timestamp and pd.to_datetime. Your performance issues here are partly due to this behaviour described in the docs:

pd.Series.dt.date returns an array of python datetime.date objects

Using object dtype in this way removes vectorisation benefits, as operations then require Python-level loops.

Use groupby operations for aggregating by date

Pandas already has functionality to group by date via normalizing time:

for day, df_day in df.groupby(df.index.floor('d')):
    df_day_t = df_day.between_time('08:30', '09:30')
    # do something

As another example, you can access a slice for a particular day in this way:

g = df.groupby(df.index.floor('d'))
my_day = pd.Timestamp('2017-01-01')
df_slice = g.get_group(my_day)