0
$\begingroup$

I have a dataframe , here above a sample :

cluster   Browser   OS   page   convert
0-3         10      11   100      10
0-3         09      10   100      10
0-3         10      11   101      50
4-7         09      10   101     100
4-7         09      10   100      20
4-7         08      10   100      30

the object of this task is to find the best way of group the records that have the value of 3 columns (3 columns except convert) are same to maximize the sum of convert .

For example there is a possibility to group these rows (group by cluster Browser OS):

0-3         10      11   100      10
0-3         10      11   101      50

and we got this result

 0-3         10      11   100, 101      60

Or maybe it is better to group by : cluster, OS and page like this :

   4-7         09      10   100      20
   4-7         08      10   100      30

to have

  4-7         08, 09      10   100      50

I am not sure if Multi-Armed Bandit can solve this kind of problem. Can you help me please to have an idea to solve it.

Thanks

$\endgroup$

1 Answer 1

0
$\begingroup$

The easiest way is to compute the "within group variation". Theoretically, within group variation should be small because subjects (rows) in the same group/cluster are similar. By comparing within group variations (smaller is better), then you can easily to tell how to group/cluster these rows.

$\endgroup$
2
  • $\begingroup$ thank you for your response, but here in my task the better groupement is the groupment which have the bigger sum of convert.. I am not sure if "within group variation" can be the solution $\endgroup$
    – user17241
    Feb 19, 2022 at 14:33
  • $\begingroup$ sum of convert is just a sum of your response variable, it does not tell how similar between subjects. For example, group A has 3 subject: 100, 101, 101.3 and group B also has three subjects: 2, 4, 3. Based on these toy data, you can tell, why we need to compute within group variation. $\endgroup$
    – DaCard
    Feb 19, 2022 at 14:42

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.