如何计算点击率

How to calculate the click-through rate

举个例子,我有这个数据;

    datetime    keyword COUNT
0   2016-01-05  a_click 100
1   2016-01-05  a_pv    200
2   2016-01-05  b_pv    150
3   2016-01-05  b_click 90
4   2016-01-05  c_pv    120
5   2016-01-05  c_click 90

我想将其转换为该数据

    datetime    keyword ctr
0   2016-01-05  a       0.5
1   2016-01-05  b       0.6
2   2016-01-05  c       0.75

我可以使用脏代码转换数据,但我想以优雅的方式进行。

你可以:

df['action'] = df.keyword.str.split('_').str.get(-1)
df['keyword'] = df.keyword.str.split('_').str.get(0)
df = df.set_index(['datetime', 'keyword', 'action']).unstack().loc[:, 'COUNT']
df['ctr'] = df.click.div(df.pv)


action              click   pv   ctr
datetime   keyword                  
2016-01-05 a          100  200  0.50
           b           90  150  0.60
           c           90  120  0.75

使用 groupby 的替代方法:

df2['key_word'] = df2.apply(lambda x: x.keyword.split('_')[0], axis=1)
df2['key_action'] = df2.apply(lambda x: x.keyword.split('_')[1], axis=1)


def compute_ctr(g):
    ctr = g[g.key_action == 'click'].COUNT.values[0] / g[g.key_action == 'pv'].COUNT.values[0]
    result = {'datetime': g.iloc[0,0], 'ctr': ctr}
    return pd.Series(result)


rslt = df2.groupby('key_word').apply(compute_ctr)
rslt.reset_index(inplace=True, drop=False)
print(rslt)


    ctr  datetime keyword
0   0.5  5/1/2016       a
1   0.6  5/1/2016       b
2  0.75  5/1/2016       c