具有不同 window_size 的多个列的移动平均值

Question

具有多维相关性的大型数据框，包括：

日期，测量 1，测量 2，...测量 10，变体

我先用这个函数计算了最近两个月变异发生的频率，然后计算了平均值。

def window_s_prüfpunkte(df, config_col, date_col, data_cols):
results = {}
for config in df[config_col].unique():
    config_df = df[df[config_col] == config]
    date_col = pd.to_datetime(config_df[date_col])
    monthly_count = config_df.groupby([date_col.dt.year, 
    date_col.dt.month]).count().rename_axis(['year', 'month'])[data_cols].reset_index()
    quantity_average = round(monthly_count[data_cols].mean().mean())
    results[config] = {'monthly_count': monthly_count,
                       'quantity_average': quantity_average}
return results

unique_configs = FSC3_Z_df_12M['Configuration Summary'].unique()
window_results = {}
for config in unique_configs:
   config_df = FSC3_Z_df_12M[FSC3_Z_df_12M['Configuration Summary'] == config]
   window_results[config] = window_s_prüfpunkte(config_df, 'Configuration Summary', 
   'Datum LP', COLS_Delta_Z)

 

现在我想通过插入各个变量的平均值作为窗口大小来计算每次测量的移动平均值。

def moving_average_by_config(df, config_col, date_col, data_cols, window_results):
results = []
for config in df[config_col].unique():
    config_df = df[df[config_col] == config]
    config_window_size = window_results[config]
    for data_col in data_cols:
        config_df[f'{data_col}_MA'] = config_df[data_col].rolling(window=config_window_size).mean()
    results.append(config_df)
return pd.concat(results)

不幸的是，这种方法行不通

Answer 1

我会建议这样的事情：

# Create example dataframe
dates = pd.date_range('2022-01-01', '2022-02-28', freq='D')
variants = ['A', 'B', 'C']
data_cols = ['Measurement 1', 'Measurement 2', 'Measurement 3']
df = pd.DataFrame({
    'Date': np.repeat(dates, len(variants) * len(data_cols)),
    'Variant': np.tile(np.repeat(variants, len(data_cols)), len(dates)),
    'Measurement 1': np.random.randn(len(dates) * len(variants) * len(data_cols)),
    'Measurement 2': np.random.randn(len(dates) * len(variants) * len(data_cols)),
    'Measurement 3': np.random.randn(len(dates) * len(variants) * len(data_cols))
})
df['Date'] = pd.to_datetime(df['Date'])

def window_s_prüfpunkte(df, config_col, date_col_name, data_cols):
    results = {}
    for config in df[config_col].unique():
        config_df = df[df[config_col] == config]
        date_col = pd.to_datetime(config_df[date_col_name])
        monthly_count = config_df.groupby([date_col.dt.year, 
        date_col.dt.month]).count().rename_axis(['year', 'month'])[data_cols].reset_index()
        quantity_average = round(monthly_count[data_cols].mean().mean())
        results[config] = {data_col: quantity_average for data_col in data_cols}
    return results


def moving_average_by_config(df, config_col, date_col, data_cols, window_sizes):
    results = []
    for config in df[config_col].unique():
        config_df = df[df[config_col] == config]
        for data_col in data_cols:
            config_window_size = window_sizes[config][data_col]
            config_df[f'{data_col}_MA'] = config_df[data_col].rolling(window=config_window_size).mean()
        results.append(config_df)
    return pd.concat(results)

window_sizes = window_s_prüfpunkte(df, 'Variant', 'Date', ['Measurement 1', 'Measurement 2', 'Measurement 3'])
moving_average_df = moving_average_by_config(df, 'Variant', 'Date', ['Measurement 1', 'Measurement 2', 'Measurement 3'], window_sizes)

初始数据框是

          Date Variant  Measurement 1  Measurement 2  Measurement 3
0   2022-01-01       A      -0.598400       0.186690       0.412284
1   2022-01-01       A       1.304156      -0.242278      -0.357972
2   2022-01-01       A      -0.689837      -0.583031       0.987450
3   2022-01-01       B       0.096719       1.300683       0.681955
4   2022-01-01       B      -0.034100      -0.540658       0.878540
..         ...     ...            ...            ...            ...
526 2022-02-28       B      -1.311980       0.702785       0.427267
527 2022-02-28       B      -1.322088      -1.358726      -0.429957
528 2022-02-28       C       0.753643      -0.811415       0.347925
529 2022-02-28       C       0.889842       0.086922      -1.129692
530 2022-02-28       C      -0.424379      -1.005491      -0.362359

输出是

          Date Variant  Measurement 1  Measurement 2  Measurement 3  \
0   2022-01-01       A      -0.598400       0.186690       0.412284   
1   2022-01-01       A       1.304156      -0.242278      -0.357972   
2   2022-01-01       A      -0.689837      -0.583031       0.987450   
9   2022-01-02       A       0.640133      -1.602435       0.477864   
10  2022-01-02       A       0.489324      -0.585138       0.722814   
..         ...     ...            ...            ...            ...   
520 2022-02-27       C      -0.293254      -0.555126       0.414370   
521 2022-02-27       C      -0.428985       1.110329      -0.242367   
528 2022-02-28       C       0.753643      -0.811415       0.347925   
529 2022-02-28       C       0.889842       0.086922      -1.129692   
530 2022-02-28       C      -0.424379      -1.005491      -0.362359   

     Measurement 1_MA  Measurement 2_MA  Measurement 3_MA  
0                 NaN               NaN               NaN  
1                 NaN               NaN               NaN  
2                 NaN               NaN               NaN  
9                 NaN               NaN               NaN  
10                NaN               NaN               NaN  
..                ...               ...               ...  
520         -0.038969          0.105347          0.114437  
521         -0.044342          0.100342          0.115158  
528         -0.035779          0.089320          0.131298  
529         -0.017788          0.085785          0.103497  
530         -0.019763          0.059974          0.082724  

[531 rows x 8 columns]