在Pandas中对分组列计算时间差（timedeltas）的实现方法

No problem! Let's walk through how to compute the time differences between consecutive entries for each person in your DataFrame. Here's a step-by-step solution:

Step 1: Set Up the DataFrame #

First, we need to properly construct the DataFrame with both the person and timestamp columns (since your timestamps are tied to specific people). Let's create that:

import pandas as pd
import datetime

timestamps = [
    datetime.datetime(2018, 1, 1, 10, 0, 0, 0), # person 1
    datetime.datetime(2018, 1, 1, 10, 0, 0, 0), # person 2
    datetime.datetime(2018, 1, 1, 11, 0, 0, 0), # person 2
    datetime.datetime(2018, 1, 2, 11, 0, 0, 0), # person 2
    datetime.datetime(2018, 1, 1, 10, 0, 0, 0), # person 3
    datetime.datetime(2018, 1, 2, 11, 0, 0, 0), # person 3
    datetime.datetime(2018, 1, 4, 10, 0, 0, 0), # person 3
    datetime.datetime(2018, 1, 5, 12, 0, 0, 0) # person 3
]

# Corresponding person IDs for each timestamp
persons = [1, 2, 2, 2, 3, 3, 3, 3]

# Create the DataFrame
df = pd.DataFrame({'person': persons, 'timestamp': timestamps})

Step 2: Compute Grouped Timedeltas #

To calculate the time difference between consecutive entries for each person, we'll:

• Group the DataFrame by the person column
• Use the diff() method on the timestamp column (this calculates the difference between each row and the prior row in the group)
• Assign the result to a new column (e.g., time_since_last)

Here's the code:

# Calculate timedeltas within each person group
df['time_since_last'] = df.groupby('person')['timestamp'].diff()

Step 3: View the Result #

Let's print the resulting DataFrame to see the timedeltas:

print(df)

Output: #

person           timestamp time_since_last
0       1 2018-01-01 10:00:00             NaT
1       2 2018-01-01 10:00:00             NaT
2       2 2018-01-01 11:00:00        0 days 01:00:00
3       2 2018-01-02 11:00:00        1 days 00:00:00
4       3 2018-01-01 10:00:00             NaT
5       3 2018-01-02 11:00:00        1 days 01:00:00
6       3 2018-01-04 10:00:00        1 days 23:00:00
7       3 2018-01-05 12:00:00        1 days 02:00:00

Key Notes: #

• NaT (Not a Time) appears for the first entry of each group because there's no prior timestamp to compare against.
• The diff() method preserves the datetime type, so the result is a Timedelta object which you can further manipulate (e.g., convert to total seconds, days, etc.):

  # Convert timedelta to total hours
  df['hours_since_last'] = df['time_since_last'].dt.total_seconds() / 3600
  

内容的提问来源于stack exchange，提问作者Michael Dorner