当某列满足基于另一列的特定条件时，是否有一种方法可以迭代地找到数据帧的索引？

Question

I have a pandas dataframe with ReadTime as Index as follows:

  ReadTime    A       B
2/4/18 0:00 6008.6  6013.55
2/4/18 0:01 6008.65 6013.6
2/4/18 0:02 6009.15 6014.05
2/4/18 0:03 6014.00 6014.1
2/4/18 0:04 6009.1  6013.7
2/4/18 0:05 6008.75 6013.65
2/4/18 0:06 6008.7  6013.25
2/4/18 0:07 6008.3  6013.25
2/4/18 0:08 6015.00 6013
2/4/18 0:09 6008.3  6003.55
2/4/18 0:10 6008.65 6013.65
2/4/18 0:11 6008.75 6013.6
2/4/18 0:12 6008.7  6013.7
2/4/18 0:13 6008.65 6013.55
2/4/18 0:14 6014.00 6013.3
2/4/18 0:15 6008.6  6013.5
2/4/18 0:16 6008.55 6013.4
2/4/18 0:17 6008.55 6013.55
2/4/18 0:18 6008.65 6013.55
2/4/18 0:19 6018    6013.6

I would like to check if values in A are greater or equal to values in B iteratively and create a new dataframe with timestamp at which it happened. Repeat the analysis with timestamp in which previous condition was satisfied.

Sample Results are as follows:

  ReadTime      C
2/4/18 0:00 2/4/18 0:03
2/4/18 0:03 2/4/18 0:08
2/4/18 0:08 2/4/18 0:14
2/4/18 0:14 2/4/18 0:19

Thanks for the help in advance.

Edit: C column refers to the timestamp at which the condition was satisfied. (ie Value in A was greater than or equal to value in B considering a value in timestamp. For Eg: At 2/4/18 0:00, value of B was 6013.55. So going through the values in A after that timestamp, we can see that at 2/4/18 0:03, value of A was 6014, which exceeded the value of B (6013.55). So, 2/4/18 0:03 was brought into C corresponding to 2/4/18 0:00.

Answer 1

Here is (if I understand correctly) a solution:

import numpy as np
df['C'] = np.where(df.A > df.B, df.index, np.nan).bfill().shift(-1)
df['X'] = (df.A > df.B).cumsum()
df = df.drop_duplicates(subset=['X'], keep='first')
df = df[['C']]

First, we fill a column named C with the timestamp of the rows where the condition is met, and put NaN elsewhere. We backfill it so that all previous rows have the same timestamp (up to the one where the condition was last met). We then shift backwards by one row (to prepare for the next step).

In order to make the indices align the way you want, we need to group the rows. We can do this by combining your condition and cumsum() , which treats True as 1 and False as 0 . Now we can drop all the rows in a group (which all have the same timestamp in C ) except for the first. This should give you the output you want.

Note: your desired output doesn't match your input (at 2/4/18 0:03 , B is greater than, not less than, A ), so the answer doesn't match your example perfectly. But I think I've gotten the spirit of what you're asking - if I'm right, please correct the question, and if I'm wrong, comment and I'll change my answer.