访问Pandas.Dataframe中的邻居行

Question

I'm trying to calculate local max and min for a series of data: if current row value is greater or lower both following and preceding row, set it to current value, else set to NaN. Is there any more elegant way to do it, other than this one:

import pandas as pd
import numpy as np

rng = pd.date_range('1/1/2014', periods=10, freq='5min')
s = pd.Series([1, 2, 3, 2, 1, 2, 3, 5, 7, 4], index=rng)
df = pd.DataFrame(s, columns=['val'])
df.index.name = "dt"
df['minmax'] = np.NaN

for i in range(len(df.index)):
    if i == 0:
        continue
    if i == len(df.index) - 1:
        continue
    if df['val'][i] >= df['val'][i - 1] and df['val'][i] >= df['val'][i + 1]:
        df['minmax'][i] = df['val'][i]
        continue
    if df['val'][i] <= df['val'][i - 1] and df['val'][i] <= df['val'][i + 1]:
        df['minmax'][i] = df['val'][i]
        continue

print(df)

Result is:

                     val  minmax
dt                              
2014-01-01 00:00:00    1     NaN
2014-01-01 00:05:00    2     NaN
2014-01-01 00:10:00    3       3
2014-01-01 00:15:00    2     NaN
2014-01-01 00:20:00    1       1
2014-01-01 00:25:00    2     NaN
2014-01-01 00:30:00    3     NaN
2014-01-01 00:35:00    5     NaN
2014-01-01 00:40:00    7       7
2014-01-01 00:45:00    4     NaN

Answer 1

We can use shift and where to determine what to assign the values, importantly we have to use the bit comparators & and | when comparing series. Shift will return a Series or DataFrame shifted by 1 row (default) or the passed value.

When using where we can pass a boolean condition and the second param NaN tells it to assign this value if False .

In [81]:

df['minmax'] = df['val'].where(((df['val'] < df['val'].shift(1))&(df['val'] < df['val'].shift(-1)) | (df['val'] > df['val'].shift(1))&(df['val'] > df['val'].shift(-1))), NaN)
df
Out[81]:
                     val  minmax
dt                              
2014-01-01 00:00:00    1     NaN
2014-01-01 00:05:00    2     NaN
2014-01-01 00:10:00    3       3
2014-01-01 00:15:00    2     NaN
2014-01-01 00:20:00    1       1
2014-01-01 00:25:00    2     NaN
2014-01-01 00:30:00    3     NaN
2014-01-01 00:35:00    5     NaN
2014-01-01 00:40:00    7       7
2014-01-01 00:45:00    4     NaN