K******Q
发帖数: 62 | 1 什么统计方法能用来发现收到邮件数和unsubscription的关系？从而找到最优的邮件提
醒数目？
我能想到的是logistic regression来建立unsubscription和收到邮件数目的关系，但
是关于收到邮件数目这个variable，怎样建立？应该选哪些时间段的呢？过去3个月的
？建立模型以后怎样发现最优的邮件提醒量？
大家有什么建议尽管提啊，多谢了！ |
G**7
发帖数: 391 | 2 I think the method you mentioned is correct. There should be a variable for
邮件提醒量. Gather information for the amount of emails. |
n**m
发帖数: 156 | 3 I think dummy variable is also an option. Usually I see at most 4 options in
mail subscription.
1. immediately 2. daily 3 weekly 4 monthly
As I see, to set up 3 dummies for the variable is very intuitive and easier
to explain. |
c*****1
发帖数: 131 | 4 横坐标为邮件数，纵坐标为unsubscription，画个图，结果是什么应该是一目了然的。
为啥什么都要往model上靠,费力不多，结果还不一定准。最近在interview，和一
hiring manager谈到，工业界大多数modeling本色很容易，干个两年就是体力活了，难
的是business insight 和 good data manipulation的能力。
【在 K******Q 的大作中提到】
: 什么统计方法能用来发现收到邮件数和unsubscription的关系？从而找到最优的邮件提
: 醒数目？
: 我能想到的是logistic regression来建立unsubscription和收到邮件数目的关系，但
: 是关于收到邮件数目这个variable，怎样建立？应该选哪些时间段的呢？过去3个月的
: ？建立模型以后怎样发现最优的邮件提醒量？
: 大家有什么建议尽管提啊，多谢了！
|
K******Q
发帖数: 62 | 5 如果简单的方法能解决当然不用model了，你说纵坐标为unsubscription画图，
unsubscription就两个value：yes/no, 这样画出来的图结果怎么一目了然了？还是你
有什么高级的图可以用，能不能share？
就算用model，business sight还是很重要的，没有好的business sight 如何选择有效
的变量。但是我这个case光用business sight能给出如何控制广告邮件的时间和数量从
而降低unsubsription么？如果你有什么高见，请赐教。
【在 c*****1 的大作中提到】
: 横坐标为邮件数，纵坐标为unsubscription，画个图，结果是什么应该是一目了然的。
: 为啥什么都要往model上靠,费力不多，结果还不一定准。最近在interview，和一
: hiring manager谈到，工业界大多数modeling本色很容易，干个两年就是体力活了，难
: 的是business insight 和 good data manipulation的能力。
|
c*****1
发帖数: 131 | 6 business sight != 如何选择有效的变量
In this case, I think if you can not draw some insights from the chart, then
you will not have significant results from modeling either.
The simple, the better!
【在 K******Q 的大作中提到】
: 如果简单的方法能解决当然不用model了，你说纵坐标为unsubscription画图，
: unsubscription就两个value：yes/no, 这样画出来的图结果怎么一目了然了？还是你
: 有什么高级的图可以用，能不能share？
: 就算用model，business sight还是很重要的，没有好的business sight 如何选择有效
: 的变量。但是我这个case光用business sight能给出如何控制广告邮件的时间和数量从
: 而降低unsubsription么？如果你有什么高见，请赐教。
|
l*********s
发帖数: 5409 | 7 The truth is quite the opposite. Charts are mostly used to assist for
understanding analytic results, not to replace nor to guide modeling.
then
【在 c*****1 的大作中提到】
: business sight != 如何选择有效的变量
: In this case, I think if you can not draw some insights from the chart, then
: you will not have significant results from modeling either.
: The simple, the better!
|
c*****1
发帖数: 131 | 8 Seems that your first step in projects is always modeling without some data
exploration.
Yes, Charts are mostly used to assistt for understanding analytic results,
but it could also give you some idea in special case like this about the
data.
I did not say Charts is used to guide modeling. How many variables in this
case? if only one variable or just a few, the insight you get from the charts are
similar to what you get from modeling.
【在 l*********s 的大作中提到】
: The truth is quite the opposite. Charts are mostly used to assist for
: understanding analytic results, not to replace nor to guide modeling.
:
: then
|
l*********s
发帖数: 5409 | 9 The first is step is always data cleaning, for sure :-)
Look, charts don't lie but they do mislead. A 2D projection of high
dimension object is not an "OK" approximation. You have to take a
comprehensive view, which to be reflected in your modeling.
Finally,you don't do variable selection by chart,period. It may give you
some guesses, but they are only guesses.
You do it with your business goal/ business insights / and the help of
modeling techniques. If your company is not doing things this way, it is not a worthy place to work at.
data
charts are
【在 c*****1 的大作中提到】
: Seems that your first step in projects is always modeling without some data
: exploration.
: Yes, Charts are mostly used to assistt for understanding analytic results,
: but it could also give you some idea in special case like this about the
: data.
: I did not say Charts is used to guide modeling. How many variables in this
: case? if only one variable or just a few, the insight you get from the charts are
: similar to what you get from modeling.
|
c*****1
发帖数: 131 | 10 I agree with you to some extent.
I have been developing models for over 5 years. I am really tired of
data manipulation/variable selection/model validation/deployment ... It is like a labor work
for me right now.
For the case only have one or two variable like this, I would rather watch
the relationship between the dependent/independent variables first, then
decide the next step.
Nice discussing with you, let's stop here.
not a worthy place to work at.
【在 l*********s 的大作中提到】
: The first is step is always data cleaning, for sure :-)
: Look, charts don't lie but they do mislead. A 2D projection of high
: dimension object is not an "OK" approximation. You have to take a
: comprehensive view, which to be reflected in your modeling.
: Finally,you don't do variable selection by chart,period. It may give you
: some guesses, but they are only guesses.
: You do it with your business goal/ business insights / and the help of
: modeling techniques. If your company is not doing things this way, it is not a worthy place to work at.
:
: data
|
