twitter 电面1

pizzafan 发表于 2017-12-24 00:20
楼主你是找人内推的吗？还是海投？

就是海投

恭喜楼主，请问楼主什么背景？

phryne 发表于 2017-12-24 02:58
恭喜楼主，请问楼主什么背景？

苦逼专业phd，非物理，神经学这种哈。

Alice007 发表于 2017-12-24 02:30
贴个我现场写的：
a. if log in yesterday: 1/3 - spend more than five minutes: 4/5: 1/3 *4/5 = 4/15 ...

有点疑问，请问，为什么：如果是10分钟的话，a,c变小的scale是一样的“呢？

忘了谢谢楼主分享！

yiliaobailiao 发表于 2017-12-24 05:25
有点疑问，请问，为什么：如果是10分钟的话，a,c变小的scale是一样的“呢？

其实我统计学基本比较薄弱，就是感觉应该一样，反正我面试时候就是这么说的，面试官也表示满意

Alice007 发表于 2017-12-24 02:30
贴个我现场写的：
a. if log in yesterday: 1/3 - spend more than five minutes: 4/5: 1/3 *4/5 = 4/15 ...

hmm，这个其实就是假设了log in和时间有某种程度的独立性

"if they log in yesterday and log in today, the probability they spend more than 5 mins on their page is ⅘ "
P(t >=5 | y = login and t = login)  ＝ 4/5
"If a user did not log in yesterday and log in today, the probability they spend more than 5 mins on their page is ½"
P(t >= 5| y = not login and t = login) = 1/2

P(y = login and t = login| t>= 5) = p(t>=5|y = login and t = login)p(y = login and t = login)/p(t>=5) = (1/9*4/5)/(1/9*4/5+2/9*1/2)  = 4/9

如果是问的是大于 5 分钟 比较 make sense。 what is the probability of a user log in for both today and yesterday if we see a user spend more than 5 mins on their page today

补充内容 (2017-12-31 10:08):
不管怎样非常感谢楼主分享！

谢谢层主，同意你的解答！

这里要不要make an assumption呢？
可以假设exponential distribution, so P(today & yesterday | T > 10 | T > 5) = P(today & yesterday | T > 5).. 如果问11mins的话就解不了了吧？