一亩三分地论坛

 找回密码
 获取更多干货,去instant注册!

扫码关注一亩三分地公众号
查看: 7990|回复: 12
收起左侧

Quantitative Analyst @ Google

[复制链接] |试试Instant~ |关注本帖
chuendes 发表于 2015-3-18 05:33:42 | 显示全部楼层 |阅读模式

2015(1-3月) 分析|数据科学类 博士 全职@Google - 网上海投 - 技术电面 |Other

注册一亩三分地论坛,查看更多干货!

您需要 登录 才可以下载或查看,没有帐号?获取更多干货,去instant注册!

x
Second Round Interview.
. Waral 鍗氬鏈夋洿澶氭枃绔,
Q1. Fair/Unfair Coin
. visit 1point3acres.com for more.
mean p^
sd: s_n

CI P^ +- /sqrt(n)*t(a/2)


Python
Orange
with open () as f:
       
. 鐗涗汉浜戦泦,涓浜╀笁鍒嗗湴

. 鍥磋鎴戜滑@1point 3 acres
Q2.  Without seening the sore_A and score_B. Given you only the first column Rating of 1000 length.How do you interprete the searching engine's change is good or not. The rating is provided by human beings. Need to consider bias of the human. Many detals  discussed.
. Waral 鍗氬鏈夋洿澶氭枃绔,. more info on 1point3acres.com
data.csv:
Rating, Score_A        , Score_B
‘Good’, 100.01        , 90
‘Bad’ , 50                , 60
‘Same’, 40                , 44

Query: Britnnny SPears
E_1 | B_1
E_2 | B_2
…   | ...

Good = 3, 4 -> p1^=325 s1_n n=2000
Bad  = 5, 4 -> p2^=475 s2_n n=2000
Good - Bad = -200, -100 -> -150
Same = 200, 200 -> 200

p1^ - p2^ > 10
p2^ - p1^ > 10
(p1 - p2)^ > 10
# Good >

var(p1-p2) = val(p1) + var(p2)
s_diff = sqrt(s1_n^2 + s2_n^2)
s1_n = /n
+- s1_n*1.96/sqrt(n)


Q3. You can see the last two column now.
data.csv:
Rating, Score_A        , Score_B
‘Good’, 100.01        , 90
‘Bad’ , 50                , 60
‘Same’, 40                , 44

.1point3acres缃
The following is proposed by the interviewer to extimate the rating from scores.

Good = 1, Bad = 0, Same = .5
Rating = \alpha + \beta_1*score_a + \beta2*score_b + \epsilon
If rating >.5 -> Good
If rating <-.5 -> Bad
Else: Same


I explained what the algorithm does and provide mine as below.
p(x) = A*1/(1+exp(-x.w))+B
-google 1point3acres
. Waral 鍗氬鏈夋洿澶氭枃绔,

评分

4

查看全部评分

本帖被以下淘专辑推荐:

vo飞天小女警 发表于 2015-3-19 09:38:23 | 显示全部楼层
为什么我从第一题到最后一题都没看懂 TOT. Waral 鍗氬鏈夋洿澶氭枃绔,
楼主可以稍微解释下么么
楼主 big cong  是不是要去on site了
回复 支持 1 反对 0

使用道具 举报

 楼主| chuendes 发表于 2015-3-19 10:13:50 | 显示全部楼层
vo飞天小女警 发表于 2015-3-19 09:38
为什么我从第一题到最后一题都没看懂 TOT
楼主可以稍微解释下么么
楼主 big cong  是不是要去on site了 { ...
. more info on 1point3acres.com
Q1. Warming up.... Usual fair/unfair coins check. Z-distribution and T-distribution.
Q2. Human provide their opinions upon the change of a searching engine.: Good, Bad, Same. For example, you have 1000 response from 1000 people. Note that sample may be bias.  How to interpret this? I have applied several methods including boosting, resampling.
Q3. Follow Q2. You have scores, do regression or classification.

I am not a statistician, still waiting for the feedback. Thank you for your bless.
回复 支持 反对

使用道具 举报

superddt 发表于 2015-3-19 12:13:57 | 显示全部楼层
For Q2, quantify "Good", "Same", and "Bad" and then construct a Bootstrapping confidence interval on the mean should be the robust way of doing it.  Since we do not have controls (for example, let the raters compare the tow engines that are actually the same), it is hard to adjust for the human bias, probably need more information.

For Q3, some clarification is needed:
1) What are the meanings of Score A/B? Is Score A the score given to engine A or the score given by rater A?.1point3acres缃
2) For each row, is the score a subjective evaluation given by the same rater or an objective evaluation given by some gold standard?
回复 支持 反对

使用道具 举报

vo飞天小女警 发表于 2015-3-25 06:08:59 | 显示全部楼层
坐等 onsite 面经  好好奇博士们的题目
回复 支持 反对

使用道具 举报

vo飞天小女警 发表于 2015-3-31 12:23:52 | 显示全部楼层
楼主要去onsite 了么
回复 支持 反对

使用道具 举报

lmcshl 发表于 2015-5-20 11:05:38 | 显示全部楼层
想问问LZ后来有去onsite吗?有没有更多的题目可以分享一下?多谢!
回复 支持 反对

使用道具 举报

guojunchimacmac 发表于 2015-10-14 08:53:00 | 显示全部楼层
Q2: what is it asking? Is it like a multinomial/categorical distribution?
Q3: Would a interval for category "same" make more sense?
Rating = \alpha + \beta_1*score_a + \beta2*score_b + \epsilon
If rating > p1 -> Good
If rating < p2 -> Bad
Else (p1, p2): Same
回复 支持 反对

使用道具 举报

aiweiwei 发表于 2015-10-14 09:31:20 | 显示全部楼层
gogole怎么还有quant的职位啊,不是应该是software engineer的职位才对嘛
回复 支持 反对

使用道具 举报

jiayao0503 发表于 2016-6-12 09:40:52 | 显示全部楼层
thanks for sharing!
回复 支持 反对

使用道具 举报

ccsherry 发表于 2016-10-3 01:40:47 | 显示全部楼层
thanks for sharing
回复 支持 反对

使用道具 举报

asyz13jinage 发表于 2016-10-16 20:15:34 | 显示全部楼层
非常感谢楼主分享
回复 支持 反对

使用道具 举报

wangyuanyuan 发表于 2016-11-2 13:02:21 | 显示全部楼层
thank you for sharing! 答主好强,既能面统计,又能面cs。请问楼主到底啥背景? 膜拜!!
回复 支持 反对

使用道具 举报

本版积分规则

请点这里访问我们的新网站:一亩三分地Instant.

Instant搜索更强大,不扣积分,内容组织的更好更整洁!目前仍在beta版本,努力完善中!反馈请点这里

关闭

一亩三分地推荐上一条 /5 下一条

手机版|小黑屋|一亩三分地论坛声明 ( 沪ICP备11015994号 )

custom counter

GMT+8, 2016-12-4 01:49

Powered by Discuz! X3

© 2001-2013 Comsenz Inc. Design By HUXTeam

快速回复 返回顶部 返回列表