Pandas 每 30 分钟计算一次平均值，间隔为 +-10 分钟

编辑：下面的答案是假设每行对应一个 10 分钟的间隔。如果您有不均匀间隔的数据，我们必须手动对数据集进行 bin 以获得所需的输出：

import numpy as np

# the sampling will be computed in +/- 10 minutes from the bin
sampling_interval = np.timedelta64(10, 'm')

# get 30 minutes bins
bins_interval = "30min"
bins = df['observation_time'].dt.floor(bins_interval).unique()

avg_values = []
for grouped_bin in bins:
    # subset the dataframe in the binned intervals
    subset = df[df['observation_time'].between(
        grouped_bin - sampling_interval, 
        grouped_bin + sampling_interval
    )]
    
    avg_values.append({
        'observation_time': grouped_bin,
        'temp': subset['temp'].mean()
    })

averaged_df = pd.DataFrame(avg_values)

我不确定这是最“pythonic”的方式，但我会这样处理问题：

# we create an empty dictionary in which we'll store the computed avgs
# to turn into a DataFrame later
avg_values = []

# we iterate over the DataFrame starting at index 1 and skipping 3 rows at a time
for idx in range(1, len(df.index), 3):
    # store the observation time in a separate variable
    observation_time = df.loc[idx, 'observation_time']
    # compute the mean between the rows before the current one, the
    # current one, and the next one
    avg_in_interval = np.nanmean([
        df.loc[idx-1, 'temp'] if idx > 0 else np.nan,
        df.loc[idx, 'temp'],
        df.loc[idx+1, 'temp'] if idx < len(df.index)-1 else np.nan
    ])
    # we append the two variables to the dictionary
    avg_values.append({'observation_time': observation_time, 'temp': avg_in_interval})

# new DataFrame
averaged_df = pd.DataFrame(avg_values)

或者，以更紧凑和通用的方式，因此您可以配置平均的间隔宽度，

interval_width = 3 # assuming it is an odd number
starting_idx = interval_width // 2
avg_values = []

for idx in range(starting_idx, len(df.index), interval_width):
    avg_values.append({
        'observation_time': df.loc[idx, 'observation_time'],
        'temp': np.mean(df.iloc[idx-starting_idx:idx+starting_idx]['temp'])
    })

averaged_df = pd.DataFrame(avg_values)

你也可以把它变成一个函数来保持你的代码干净：

def get_averaged_df(df, interval_width: int):
    if interval_width % 2 == 0:
        raise Error("interval_width must be an odd integer")

    starting_idx = interval_width // 2
    avg_values = []

    for idx in range(starting_idx, len(df.index), interval_width):
        avg_values.append({
            'observation_time': df.loc[idx, 'observation_time'],
            'temp': np.mean(df.iloc[idx-starting_idx:idx+starting_idx]['temp'])
        })

    return pd.DataFrame(avg_values)


averaged_df = get_averaged_df(df, 3)