python - 按组查看趋势变化(python pandas 数据框)

Question

我正在尝试根据下面现有的数据框创建一个新的数据框。我的目标是计算点击次数的平均变化并相应地对事件进行分类。

现有数据框 df:

campaign    |     date     |    clicks 
   A           2015-10-11       255
   A           2015-10-12       367
   A           2015-10-13       489
   B           2015-10-11       500
   B           2015-10-15       122
   C           2015-10-11       33

目标数据框 df_categorized:

campaign | avg_change |   category
   A        0.3858       increasing
   B        -0.756       decreasing
   C           0         no change

我试过这段代码，但收到错误消息 TypeError: 'long' object does not support item assignment

 #standard packages
 import pandas as pd
 import numpy as np

 #upload data into df
 df = pd.read_csv('C:\Users\xxx\Documents\\ad_table.csv')

 df.head()
 campaign    |     date     |    clicks 
   A           2015-10-11       255
   A           2015-10-12       367
   A           2015-10-13       489
   B           2015-10-11       500
   B           2015-10-15       122
   C           2015-10-11       33

 #create empty dataframe
 columns = ['group','avg_change', 'category']     
 df_categorized = pd.DataFrame(columns=columns)

 df_categorized['avg change'] = df.clicks.apply(lambda df:       df.pct_change().abs().mean())

 #create column
 df_categorized['category'] = 0
 # going up
 df_categorized['category'][df_categorized['avg change'] > 0] = "increasing"
 # going down
 df_categorized['category'][df_categorized['avg change'] < 0] = "decreasing"
 #no change
 df_categorized['category'][df_categorized['avg change'] = 0] = "no change"

Answer 1

您可以 groupby在“事件”上，然后 apply计算 pct_change 的 lambda并返回 mean .然后您可以在此reset_index 并使用np.where 添加额外的类别列。 :

In [239]:
gp = df.groupby('campaign')['clicks'].apply(lambda x: x.pct_change().mean()).reset_index(name='avg_change').fillna(0)
gp['category'] = np.where(gp['avg_change'] < 0, 'decreasing', np.where(gp['avg_change'] > 0, 'increasing', 'no change'))
gp

Out[239]:
  campaign  avg_change    category
0        A     0.38582  increasing
1        B    -0.75600  decreasing
2        C     0.00000   no change

这个:

df_categorized['avg change'] = df.clicks.apply(lambda df: df.pct_change().abs().mean())

不会工作，你在列上调用 apply 所以 lambda 将是每个行元素，在这种情况下是一个 int 因此你得到错误:

AttributeError: 'int' object has no attribute 'pct_change'

即使没有这个，它也不会为您提供每个广告系列的 pct_change。

也不要像这样对你的 df 进行链式调用:

df_categorized['category'][df_categorized['avg change'] > 0] = "increasing"

应该是:

df_categorized.loc[df_categorized['avg change'] > 0, 'category'] = "increasing"

参见 docs