小编Jos*_*osh的帖子

我在这方面远远超出了我的联盟，所以我希望有人能指出我正确的方向。我认为这是一个优化问题，但我对 scipy.optimize 及其与纸浆的匹配方式感到困惑。此外，矩阵数学让我难以置信。因此，这个问题确实让我放慢了速度，而无需问。

问题陈述：我有一个客户数据集。对于每个客户，我可以选择 3 个选项，或者一个都不选。所以有4个选项。同样对于每个客户，我都有一个数字分数，说明每个选择的“好”程度。您可以将这个值想象为the probability of the Choice to create a future sale。

# fake data for the internet
data = {'customerid':[101,102,103,104,105,106,107,108,109,110],
        'prob_CHOICEA':[0.00317,0.00629,0.00242,0.00253,0.00421,0.00414,0.00739,0.00549,0.00658,0.00852],
        'prob_CHOICEB':[0.061,0.087,0.055,0.027,0.022,0.094,0.099,0.072,0.018,0.052],
        'prob_CHOICEC':[0.024,0.013,0.091,0.047,0.071,0.077,0.067,0.046,0.077,0.044]
       } 

# Creates pandas DataFrame 
df = pd.DataFrame(data) 
df = df.reset_index(drop=True).set_index(['customerid'])

+------------+--------------+--------------+--------------+
| customerid | prob_CHOICEA | prob_CHOICEB | prob_CHOICEC |
+------------+--------------+--------------+--------------+
|        101 |      0.00317 |        0.061 |        0.024 |
|        102 |      0.00629 |        0.087 |        0.013 |
|        103 |      0.00242 |        0.055 |        0.091 |
| …

我正在尝试将假数据插入到该表中。它不能完全随机，因为行需要有意义。下面我来解释一下。

我的数据如下所示：

每个 AcctID 可以出现多次，但最容易解释我在做什么，仅举一个 AcctID 出现两次的例子：