比较不同分辨率的 pandas 时间戳

Question

我有两个时间序列数据框（约 45k 行与 5 行）。一个具有低至毫秒的时间戳，另一个低至秒。我想在较大的数据框中创建一个新列，以便： a）一个值被附加到较大数据帧中的行，其时间戳最接近（以秒为单位）较小数据帧中的时间戳 b) 任何其他时间戳的 NaN。

larger df = 
            timestamp           price
0       2018-04-24 06:01:02.600 1
1       2018-04-24 06:01:02.600 1
2       2018-04-24 06:01:02.600 2
3       2018-04-24 06:01:02.600 4
4       2018-04-24 06:01:02.775 2
5       2018-04-24 06:01:02.825 3
6       2018-04-24 06:01:03.050 5
7       2018-04-24 06:01:03.125 6
8       2018-04-24 06:01:03.275 7
9       2018-04-24 06:01:03.300 4
10      2018-04-24 06:01:03.300 3
11      2018-04-24 06:01:03.950 5
12      2018-04-24 06:01:04.050 5


smaller df = 
   timestamp           price
0   24/04/2018 06:01:02 2
1   24/04/2018 12:33:37 4   
2   24/04/2018 14:29:34 5   
3   24/04/2018 15:02:50 6   
4   24/04/2018 15:20:04 7   

desired df =

            timestamp       price  newCol
0       2018-04-24 06:01:02.600 1   aValue
1       2018-04-24 06:01:02.600 1   NaN
2       2018-04-24 06:01:02.600 2   NaN
3       2018-04-24 06:01:02.600 4   NaN
4       2018-04-24 06:01:02.775 2   NaN
5       2018-04-24 06:01:02.825 3   NaN
6       2018-04-24 06:01:03.050 5   NaN
7       2018-04-24 06:01:03.125 6   NaN
8       2018-04-24 06:01:03.275 7   NaN
9       2018-04-24 06:01:03.300 4   NaN
10      2018-04-24 06:01:03.300 3   NaN
11      2018-04-24 06:01:03.950 5   NaN
12      2018-04-24 06:01:04.050 5   NaN

我们将非常感谢您的帮助。一般来说，我对编程还是太陌生，无法轻松解决这个问题。

非常感谢

Answer 1

reindex

为了只使用一次值，我必须从较小的数据帧中跟踪时间戳。所以我在reindex 和'nearest' 时包含了这些值。然后我在掩码中使用duplicated。

df_small_new = df_small.set_index('timestamp', drop=False)
df_small_new = df_small_new.reindex(df_large.timestamp, method='nearest')


df_large.assign(
    newcol=df_small_new.price.mask(df_small_new.timestamp.duplicated()).values)

                 timestamp  price  newcol
0  2018-04-24 06:01:02.600      1     2.0
1  2018-04-24 06:01:02.600      1     NaN
2  2018-04-24 06:01:02.600      2     NaN
3  2018-04-24 06:01:02.600      4     NaN
4  2018-04-24 06:01:02.775      2     NaN
5  2018-04-24 06:01:02.825      3     NaN
6  2018-04-24 06:01:03.050      5     NaN
7  2018-04-24 06:01:03.125      6     NaN
8  2018-04-24 06:01:03.275      7     NaN
9  2018-04-24 06:01:03.300      4     NaN
10 2018-04-24 06:01:03.300      3     NaN
11 2018-04-24 06:01:03.950      5     NaN
12 2018-04-24 06:01:04.050      5     NaN

---

pandas.merge_asof

• 重命名小数据框中的'price' 列
• 确保将direction 设置为'nearest'
• 这几乎回答了问题

---

pd.merge_asof(
    df_large,
    df_small.rename(columns={'price': 'newcol'}),
    on='timestamp', direction='nearest'
)

                 timestamp  price  newcol
0  2018-04-24 06:01:02.600      1       2
1  2018-04-24 06:01:02.600      1       2
2  2018-04-24 06:01:02.600      2       2
3  2018-04-24 06:01:02.600      4       2
4  2018-04-24 06:01:02.775      2       2
5  2018-04-24 06:01:02.825      3       2
6  2018-04-24 06:01:03.050      5       2
7  2018-04-24 06:01:03.125      6       2
8  2018-04-24 06:01:03.275      7       2
9  2018-04-24 06:01:03.300      4       2
10 2018-04-24 06:01:03.300      3       2
11 2018-04-24 06:01:03.950      5       2
12 2018-04-24 06:01:04.050      5       2