DataFrame在日期范围上重新采样 [英] DataFrame resample on date ranges

问题描述

我有一个包含列的DataFrame 开始时间"(日期时间)，结束时间"(日期时间)，模式"和其他一些列. 该表的不同行的范围内没有重叠.

I have a DataFrame that has the columns 'start_time' (datetime), 'end_time' (datetime), 'mode' and some other columns. There is no overlap in the ranges of different rows of the table.

我想创建一个新的DataFrame，对原始DataFrame的每一行进行重新采样，如下所示: 'current_time'，'mode'，其他列

I would like to create a new DataFrame, that resamples each row of the original DataFrame like so: 'current_time', 'mode', other columns

"current_time"是在原始"start_time"和"end_time"之间以给定频率进行的重采样，而所有其他列只是原始表中值的副本.

Where 'current_time' is a resample between original 'start_time' and 'end_time' with a given frequency, and all other columns are just copies of values from the original table.

示例: 原始DataFrame:

Example: original DataFrame:

             start_time                   end_time   mode
2017-06-01 06:38:00.000 2017-06-01    06:39:00.000      x
2017-06-01 17:22:00.000 2017-06-01    17:22:30.000      y

对于给定的"10S"频率，我想获取以下DataFrame:

For a given 'freq' of '10S', I'd like to get the following DataFrame:

           current_time     mode
2017-06-01 06:38:00.000        x
2017-06-01 06:38:10.000        x
2017-06-01 06:38:20.000        x
2017-06-01 06:38:30.000        x
2017-06-01 06:38:40.000        x
2017-06-01 06:38:50.000        x
2017-06-01 17:22:00.000        y
2017-06-01 17:22:10.000        y
2017-06-01 17:22:20.000        y

我正在寻找一种合理有效且优雅的方法.

I'm looking for a reasonably efficient and elegant way of doing this.

非常感谢！

推荐答案

您可以使用:

#convert columns to datetimes if necessary
df['start_time']= pd.to_datetime(df['start_time'])
df['end_time']= pd.to_datetime(df['end_time'])
#subtract 10s for no last row from values from end_time column
df['end_time']= df['end_time'] - pd.Timedelta(10, unit='s')

---

#loop by list comprehension for list of date ranges
#concat to one big DataFrame
df1 = (pd.concat([pd.Series(r.Index, 
                           pd.date_range(r.start_time, r.end_time, freq='10S')) 
                           for r in df.itertuples()])
        .reset_index())
df1.columns = ['current_time','idx']
print (df1)
         current_time  idx
0 2017-06-01 06:38:00    0
1 2017-06-01 06:38:10    0
2 2017-06-01 06:38:20    0
3 2017-06-01 06:38:30    0
4 2017-06-01 06:38:40    0
5 2017-06-01 06:38:50    0
6 2017-06-01 17:22:00    1
7 2017-06-01 17:22:10    1
8 2017-06-01 17:22:20    1

根据OP的评论进行

如果使用参数closed=left:

pd.date_range(r.start_time, r.end_time, freq='10S', closed='left')

那么可以省略减法.

#join all another columns by index
df2 = df1.set_index('idx').join(df.drop(['start_time','end_time'], 1)).reset_index(drop=True)
print (df2)
         current_time mode
0 2017-06-01 06:38:00    x
1 2017-06-01 06:38:10    x
2 2017-06-01 06:38:20    x
3 2017-06-01 06:38:30    x
4 2017-06-01 06:38:40    x
5 2017-06-01 06:38:50    x
6 2017-06-01 17:22:00    y
7 2017-06-01 17:22:10    y
8 2017-06-01 17:22:20    y

另一种解决方案:

#create column from index for last join (index values has to be unique)
df = df.reset_index()
#reshape dates to datetimeindex
df1 = (df.melt(df.columns.difference(['start_time','end_time']),
              ['start_time', 'end_time'],
              value_name='current_time')
        .drop('variable', 1)
        .set_index('current_time'))
print (df1)
                     index mode
current_time                   
2017-06-01 06:38:00      0    x
2017-06-01 17:22:00      1    y
2017-06-01 06:38:50      0    x
2017-06-01 17:22:20      1    y

#group by index column and resample, NaNs are replaced by forward filling
df2 = df1.groupby('index').resample('10S').ffill().reset_index(0, drop=True).drop('index', 1)
print (df2)
                    mode
current_time            
2017-06-01 06:38:00    x
2017-06-01 06:38:10    x
2017-06-01 06:38:20    x
2017-06-01 06:38:30    x
2017-06-01 06:38:40    x
2017-06-01 06:38:50    x
2017-06-01 17:22:00    y
2017-06-01 17:22:10    y
2017-06-01 17:22:20    y

这篇关于DataFrame在日期范围上重新采样的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！