Pandas 熊猫时间序列回顾

有一个带有时间序列的数据帧：

import pandas as pd
import numpy as np

df = pd.DataFrame({'times': np.array(['1994-07-25 15:00:00.000',
                            '1994-07-25 16:00:00.000', 
                            '1994-07-26 18:45:00.000', 
                            '1994-07-27 15:15:00.000', 
                            '1994-07-27 16:00:00.000', 
                            '1994-07-28 18:45:00.000', 
                            '1994-07-28 19:15:00.000', 
                                     ], dtype='datetime64'),
                   'diff': [0.0,0.03,0.04,0.05,0,0.01,0.0,]})

差值（两个信号之间）恢复为零，我想找出它偏离直线的时间-即非零->低于所需结果的时间

df['deviation_time_delta'] = pd.to_timedelta(['nan',
                            'nan', 
                            'nan', 
                            'nan', 
                            '2 days 00:00:00.000', 
                            'nan', 
                            '0 days 00:30:00.000', 
                                     ])

我已经试过了，但它并不漂亮，而且对任意长度偏差也不起作用：

df['diff_1'] = df['diff'].shift(1)
df['diff_2'] = df['diff'].shift(2)
df['diff_3'] = df['diff'].shift(3)
df['diff_4'] = df['diff'].shift(4)
df['times_1'] = df['times'].shift(1)
df['times_2'] = df['times'].shift(2)
df['times_3'] = df['times'].shift(3)
df['times_4'] = df['times'].shift(4)

def calc_dev_time_delta(cur_diff, diff_1, diff_2, diff_3, diff_4, cur_time, time_1, time_2, time_3, time_4):
    if cur_diff != 0.0: return np.nan
    if diff_1 == 0.0: return np.nan
    if diff_2 == 0.0: return cur_time - time_1 
    if diff_3 == 0.0: return cur_time - time_2
    if diff_4 == 0.0: return cur_time - time_3

df['dev_time_delta'] = df.apply(lambda row: calc_dev_time_delta(row['diff'], row['diff_1'], row['diff_2'],row['diff_3'],row['diff_4'], row['times'], row['times_1'], row['times_2'], row['times_3'], row['times_4']), axis=1)

---

您知道实现此结果的更好/更干净的方法吗？

如果我理解正确，您希望计算与上一行的差异，其中

diff

为0

使用

groupby

和

diff

df
   diff               times
0  0.00 1994-07-25 15:00:00
1  0.03 1994-07-25 16:00:00
2  0.04 1994-07-26 18:45:00
3  0.05 1994-07-27 15:15:00
4  0.00 1994-07-27 16:00:00
5  0.01 1994-07-28 18:45:00
6  0.00 1994-07-28 19:15:00

df['deviation_time_delta'] = df.groupby('diff')['times'].diff()
df['deviation_time_delta'].loc[df['diff']!=0] = 0
df
   diff               times deviation_time_delta
0  0.00 1994-07-25 15:00:00                  NaT
1  0.03 1994-07-25 16:00:00                    0
2  0.04 1994-07-26 18:45:00                    0
3  0.05 1994-07-27 15:15:00                    0
4  0.00 1994-07-27 16:00:00      2 days 01:00:00
5  0.01 1994-07-28 18:45:00                    0
6  0.00 1994-07-28 19:15:00      1 days 03:15:00

---

我不确定我是否明白你想要什么，但这不就是工作吗

dfZero = df[df['diff'] == 0]
dfZero['deltaT'] = dfZero.times.diff()
df = df.merge(dfZero, how='left')

print(df)

输出：

  times                diff             deltaT
0 1994-07-25 15:00:00  0.00             NaT
1 1994-07-25 16:00:00  0.03             NaT
2 1994-07-26 18:45:00  0.04             NaT
3 1994-07-27 15:15:00  0.05             NaT
4 1994-07-27 16:00:00  0.00 2 days 01:00:00
5 1994-07-28 18:45:00  0.01             NaT
6 1994-07-28 19:15:00  0.00 1 days 03:15:00

---

受godot回答和评论的启发，请参见下面我最终得出的解决方案：

df['diff_1'] = df['diff'].shift(1)

def keep_row(cur_diff, prev_diff):
    return cur_diff == 0.0 or prev_diff == 0.0

df['keep'] = df.apply(lambda row: keep_row(row['diff'], row['diff_1']), axis=1)
df_short = df[df['keep']]

df_short = df_short.drop(['diff_1'], axis=1)
df_short['diff_1'] = df_short['diff'].shift(1)
df_short['times_1'] = df_short['times'].shift(1)

def calc_deviation_time(cur_diff, prev_time, cur_time):
    if cur_diff != 0.0: return np.nan
    return cur_time - prev_time

df_short['deviation_time'] = df_short.apply( lambda row: calc_deviation_time(row['diff'], row['times_1'], row['times']), axis=1)
df_short = df_short.drop(['keep', 'diff_1', 'times_1'], axis=1)
df_short

---

谢谢你看。有一个细微的差别，它测量了零之间的时间，我在寻找零之后的第一个偏差值和随后的零之间的时间。我将试用你建议的一种变体。从times创建第二列，但移位为1，然后在新列和原始“times”列之间进行差异。。。