數據幀的值第一個實例中

Question

柱

我有DF：

     Voltage 
01-02-2017 00:00  13.1 
01-02-2017 00:01  13.2 
01-02-2017 00:02  13.3 
01-02-2017 00:03  14.1 
01-02-2017 00:04  14.3 
01-02-2017 00:04  13.5 

我想的時間：第一個實例的（HH毫米）時在電壓柱> = 14.0的值。 「充電時間」欄中應該只有一個時間值。

     Voltage Time of Full Charge 
01-02-2017 00:00  13.1 
01-02-2017 00:01  13.2 
01-02-2017 00:02  13.3 
01-02-2017 00:03  14.1   00:03 
01-02-2017 00:04  14.3 
01-02-2017 00:04  13.5 

我想沿着這些線路的東西，但不能弄明白：

df.index = pd.to_datetime(df.index) 
df.['Time of Full Charge'] = np.where(df.['Voltage'] >= 14.0), (df.index.hour:df.index.minute))  

Answer 1

需要idxmax通過條件第一指標值，不僅是必要的指數必須是唯一的：

idx = (df['Voltage'] >= 14.0).idxmax() 
df.loc[mask, 'Time of Full Charge'] = mask.idxmax().strftime('%H:%M') 
print (df) 
        Voltage Time of Full Charge 
2017-01-02 00:00:00  13.1     NaN 
2017-01-02 00:01:00  13.2     NaN 
2017-01-02 00:02:00  13.3     NaN 
2017-01-02 00:03:00  14.1    00:03 
2017-01-02 00:04:00  14.3     NaN 
2017-01-02 00:04:00  13.5     NaN 

或者：

idx = (df['Voltage'] >= 14.0).idxmax() 
df['Time of Full Charge'] = np.where(df.index == idx, idx.strftime('%H:%M'), '') 
print (df) 
        Voltage Time of Full Charge 
2017-01-02 00:00:00  13.1      
2017-01-02 00:01:00  13.2      
2017-01-02 00:02:00  13.3      
2017-01-02 00:03:00  14.1    00:03 
2017-01-02 00:04:00  14.3      
2017-01-02 00:04:00  13.5  

對於非唯一索引，可以使用MultiIndex：

df.index = [np.arange(len(df.index)), df.index] 

idx = (df['Voltage'] >= 14.0).idxmax() 
df['Time of Full Charge'] = np.where(df.index.get_level_values(0) == idx[0], 
            idx[1].strftime('%H:%M'), 
            '') 

df.index = df.index.droplevel(0) 
print (df) 
        Voltage Time of Full Charge 
2017-01-02 00:00:00  13.1      
2017-01-02 00:01:00  13.2      
2017-01-02 00:02:00  13.3      
2017-01-02 00:03:00  14.1    00:03 
2017-01-02 00:04:00  14.3      
2017-01-02 00:04:00  13.5      

Answer 2

您可以使用numpy.searchsorted()如果Voltage列進行排序：

In [260]: df.index[np.searchsorted(df.Voltage, 14)] 
Out[260]: DatetimeIndex(['2017-01-02 00:03:00'], dtype='datetime64[ns]', freq=None)