Question

我有一个如下所示的数据框：

 START_TIME   END_TIME     TRIAL_No        itemnr
 2403950      2413067      Trial: 1        P14
 2413378      2422499      Trial: 2        P03
 2422814      2431931      Trial: 3        P13
 2432246      2441363      Trial: 4        P02
 2523540      2541257      Trial: 5        P11
 2541864      2560297      Trial: 6        P10
 2560916      2577249      Trial: 7        P05

桌子一直在继续。 START_TIME和END_TIME都以毫秒为单位，这是试验的开始和结束时间。所以我想要做的是，我想将START_TIME重新采样到100毫秒bin itme并在每个START_TIME和END_TIME之间插入变量（TRIAL_No和itemnr）。在这些区域之外，这些变量的值应为“NA”。例如，对于第一行，START_TIME是2403950，END_TIME是2413067.它们之间的差异是9117毫秒。因此，“试验：1”保持9117毫秒，这是因为每个箱时间相隔100毫秒，因此需要91个箱时间。所以我想在结果数据帧中重复“Trial_1”和“P14”91次。其余部分也是如此。看起来如下：

Bin_time     TRIAL_No    itemnr
2403950      Trial: 1    P14
2404050      Trial: 1    P14
2404150      Trial: 1    P14
            ...
2413050      Trial: 1    P14
2413150      Trial: 2    P03
2413250      Trial: 2    P03

等等。我不确定是否可以直接在熊猫中进行，或者需要进行一些预处理。

Answer 1

按concat数据框创建新数据框后，我可以按行对其进行分组，并在每个组上应用resample（使用方法ffill转发填充）。

print df
#   START_TIME  END_TIME  TRIAL_No itemnr
#0     2403950   2413067  Trial: 1    P14
#1     2413378   2422499  Trial: 2    P03
#2     2422814   2431931  Trial: 3    P13
#3     2432246   2441363  Trial: 4    P02
#4     2523540   2541257  Trial: 5    P11
#5     2541864   2560297  Trial: 6    P10
#6     2560916   2577249  Trial: 7    P05

#PREDPROCESSING
#helper column for matching start and end rows
df['row'] = range(len(df))

#reshape to df - every row two times repeated for each date of START_TIME and END_TIME
starts = df[['START_TIME','TRIAL_No','itemnr','row']].rename(columns={'START_TIME':'Bin_time'})
ends = df[['END_TIME','TRIAL_No','itemnr','row']].rename(columns={'END_TIME':'Bin_time'})
df = pd.concat([starts, ends])
df = df.set_index('row', drop=True)
df = df.sort_index()

#convert miliseconds to timedelta for resampling by time 100ms
df['Bin_time'] = df['Bin_time'].astype('timedelta64[ms]')

print df
#           Bin_time  TRIAL_No itemnr
#row                                 
#0   00:40:03.950000  Trial: 1    P14
#0   00:40:13.067000  Trial: 1    P14
#1   00:40:13.378000  Trial: 2    P03
#1   00:40:22.499000  Trial: 2    P03
#2   00:40:22.814000  Trial: 3    P13
#2   00:40:31.931000  Trial: 3    P13
#3   00:40:32.246000  Trial: 4    P02
#3   00:40:41.363000  Trial: 4    P02
#4   00:42:03.540000  Trial: 5    P11
#4   00:42:21.257000  Trial: 5    P11
#5   00:42:21.864000  Trial: 6    P10
#5   00:42:40.297000  Trial: 6    P10
#6   00:42:40.916000  Trial: 7    P05
#6   00:42:57.249000  Trial: 7    P05

print df.dtypes
#Bin_time    timedelta64[ms]
#TRIAL_No             object
#itemnr               object
#dtype: object

#resample and fill missing data 
df = df.groupby(df.index).apply(lambda x: x.set_index('Bin_time').resample('100ms',how='first',fill_method='ffill'))

df = df.reset_index()
df = df.drop(['row'], axis=1)

#convert timedelta to integer back
df['Bin_time'] = (df['Bin_time'] / np.timedelta64(1, 'ms')).astype(int)

print df.head()
#  Bin_time  TRIAL_No itemnr
#0  2403950  Trial: 1    P14
#1  2404050  Trial: 1    P14
#2  2404150  Trial: 1    P14
#3  2404250  Trial: 1    P14
#4  2404350  Trial: 1    P14

编辑：

如果您希望在群组之外获取NaN，则可以在groupby之后更改代码：

#resample and fill missing data 
df = df.groupby(df.index).apply(lambda x: x.set_index('Bin_time').resample('100ms', how='first',fill_method='ffill'))

#reset only first level - drop index row
df = df.reset_index(level=0, drop=True)
#resample by 100ms, outside are NaN
df = df.resample('100ms', how='first')
df = df.reset_index()
#convert timedelta to integer back
df['Bin_time'] = (df['Bin_time'] / np.timedelta64(1, 'ms')).astype(int)

print df

重新采样开始和结束日期列

1 个答案: