Chicago – Product Development

Do you know how to make C++ programming? Or do you have relevant working experience in C++ developing?

—ABBRobot

​Sample Interview Question

Behavior Question

这些问题是我在面试中比较常见到的一些behavior questions,最好的办法就是能够把每一道题都自己过一遍,举小例子可以灵活的套用不同的问题。遵行star原则(situation, task, action, and result)整理自己的回答。最后就是求米! 谢谢大家打赏!

o        Basic attention to details:

  "        Can you tell me a bit about your current role?

  "        How would your current or former colleagues describe you?

                 "        Good Answer: hard working, meticulous, persistent

  "        Do you think you have attention to detail?

                 "        Good Answer: Yes, with examples that demonstrate understanding and depth

                  "        Bad Answer: No or wishy washy answer with lack of understanding/depth

  "        Can you give me a couple examples where your attention to detail comes into play in your current role? (ask a few follow up Qs based on responses)

                  "        Good Answer: Organization, confirming every detail, proofing and double-checking, taking time to ensure accuracy

                  "        Bad answer: Examples that don't display attn to detail

  "        How do you stay organized when you have many competing priorities?

                 "        Good Answer: task management, making a hit list, taking projects one at a time, logical approach

                 "        Bad Answer: Lack of logic, expression of overwhelm when work load is seemingly light

Desire to solve problems:

  "        Can you tell me about a time when you ran into a roadblock at work? What did you do and what was the outcome?

                 "        Good Answer: Problem solved, tried different approaches, persistence, asked questions

                 "        Bad Answer: Apprehension to ask Qs, frustration, blaming others

  "        Can you tell me about a time that you felt really successful or accomplished at work?

                 "        Good Answer: Achieving a team goal, giving others credit where due, learned something

                 "        Bad Answer: No example, selfish focus

  "        Can you tell me about a time when you were really frustrated at work?

                 "        Good Answer: genuine example, worked through

                 "        Bad Answer: blaming others, eye-rolling

o        Company culture and role fit:

  "        What excites you about our company's business and this role?

                 "        Good Answer: interest in the business, passion for improving/growing our business, desire to learn

                 "        Bad Answer: lack of excitement, excitement for responsibilities NOT included in the role

"        In what you've learned so far about the role, do you have additional questions you'd like to ask?

"        What do you want me to know about you that I might not have asked?

  Tell me about a time you have been stressed/overwhelmed.

                 "        Good answer: demonstrates self awareness.

                 "        Bad answer: stressed by too much work, ambiguity, lack of strict guidance (e.g. written docs).

o        What was a project you had where instruction was unclear? How did you make decisions in this situation?

                 "        Good answer: shows logic and thoughtful approach.

                 "        Bad answer: can't verbalize thought process. Decisions were arbitrary, lacked logic.

o        When have you had the most work to juggle? i.e. what are all the things you had to do. What factors helped you decide what to prioritize?

                 "        Good answer: large workload, logical approach.

                 "        Bad answer: intentional decision making is not apparent. Lack of logic. Workload is relatively small.

o        Tell me about a time you made a difficult decision not to do something. Why was it hard?

                 "        Good answer: Demonstrates what is logical/practical over what is desired.

                 "        Bad answer: Conveys combativeness, lack of flexibility. Irrational.

o        Tell me about a time you created a way to do something much more efficiently than it had been done before.

                 "        Good answer: Novel, impressive idea. Demonstrates inherent seeking for improvement.

                 "        Bad answer: Non answer, not novel.

o        What is something that really impresses you about your current/last company?

                 "        Good answer: Involves sharecropper's value (e.g. impactful work). Shows genuine interest.

                 "        Bad answer: Is surface level, difficult to think of, negative.

o        Self-awareness: Tell me about a time you disagreed with a team member

                 "        Great answer: Listened, tried to see it from their perspective, tried to get a 3rd party's perspective, tried to appeal to common goals, etc.  

                 "        Bad answer: I just overruled the person, I did it anyway, I did it myself

o        Communication: Tell me about a time you received feedback from a team member or colleague that you disagreed with.  

                 "        Great answer: Adjusted my style. Behavior or communication change. Try to get to root cause of the issue. Think about implications of what they are saying. Talked to them.

                 "        Bad answer: Ignored it. Overruled them and went to their / my manager. Did not talk to them about it.

概率

楼主最近在准备DS面试,其中经典的概率问题总结如下,跪求大家加米。

先从池塘里面捕捉1000条鱼,并且在它们的尾巴上做上标记,放回。一段时间后,再从池塘里面捕捉800条鱼,发现了80条做了标记的鱼。问池塘里面一共有多少条鱼?

•        由于在第二次捕捉中,标记好了的鱼占10%,那么1000条鱼就占了总数的10%,答案是10000.

-baidu 1point3acres

抛一个公正的硬币,直到正面反面都出现过为止,问抛硬币的期望次数。

•        E=2/2+2/1=3

抛一个公正的骰子,直到集齐六面为止,问抛骰子的期望次数。

可以看做是一串几个分布。在扔第一次之前,得到一个新数字的概率是p1=6/6,因为不管出现几,这个数都是之前没有出现过的。之后要得到第二个出现的数字的概率就变成了5/6。同理,要得到第k个新出现的数字的概率就是pk=(7-k)/6。根据几个分布,我们知道,得到一次成功所需要的实验的次数是1/p。所以,出现全部6个数字,我们所需要扔骰子的次数的期望是:

6/6+6/5+6/4+6/3+6/2+6/1=14.7

老师想用问卷的方式调查有多少学生作弊,问卷上只能填yes or no,并且老师给学生一枚硬币。如果学生第一次投硬币得到了Head,那他就根据自己的真实情况填是否作弊,如果是tail,就再投一次,第二次是head的话就一定会填Yes,tail的话就一定填No,最后根据问卷上的yes/No的个数,估算真正作弊的人。

•        H -> Y/N: 1/2. 1point3acres

•        T->T ->N: (1/2)*(1/2)

•        T->H ->Y: (1/2)*(1/2)

•        total number is N

•        total Yes is K

•        Total cheat is X

•        cheat rate is p=X/N

•        (p*(1/2)+p*(1/4)+(1-p)*(1/4))*N=K. check 1point3acres for more.

•        p*(1/2)+(1/4)=K/N

•        2p+1=4K/N

•        X=2K-N/2

一共有N个conference room from NO.1 to No.N。有k个meeting独立随机分配到这N个conference room。现在已知1号conference room里面被shcedule了一个meeting,问1号conference room里面被schedule的总共的meeting的数量(已知1号房存在一个meeting,也就是1号房不为空。在这个条件下求1号房总的meeting数的期望。把meeting的集合写成M1,M2,…MK,对任意的i,利用bayes公式计算条件概率). 1point3acres

•        let check all the requirement of binomial distribution are valid:

o        Trials are independent (because we can schedule a meeting in any room irrespective of whether they have meetings scheduled or not)

o        Fixed number of trials (k)

o        P(success) is the same across trials (yes, this is 1/N for every meeting assignment we have to do)

•        probability of Room 1 is not empty as event A

P(A) = 1-(1-1/N)^k

•        probability of meeting i in Rooms 1 as event B

P(B) =1/N

•        conditional probability of meeting i in room 1 is

P(B | A) = P(A | B) * P(B) / P(A) = 1* P(B) / P(A) = 1/N / (1-(1-1/N)^k)

•        the expected number of meeting in room 1 is

n = k * P(B | A) =k/N / (1-(1-1/N)^k)

GOOGLE

Timeline:

10月被reach out

11月得到JD

12月店面

1月得到店面结果

2月VO

昨天告知结果L3过(没有HC)L4(Product sense不够)

狗家的时间线是真的长,过程也是比较复杂,第一次Go through这个流程,还是觉得挺累的。因为全签了NDA所以不好细说。

大概店面:Python考了基础的Coding,考了Data structure,考了SQL,考了stats还有Product sense。本来计划的45分钟直接面了1小时25分钟。然后Feedback L3/L4都可但是stats不扎实(被“推荐”给另一个组)

VO面了3轮(有一轮因为面试官之间的沟通问题不纳入Feedback,这有伏笔后面讲)第一轮SQL,stakeholder management还有一部分BQ。Feedback:Lean L4

第二轮:ERD(因为面试题不在JD考察范围,跟Recruiter投诉之后作废)

第三轮:BQ,product sense。Feedback:down level to L3

总体来说不难,就是面试题真的非常随机,所以准备起来很难面面俱到。建议如果有兴趣狗家多了解下狗的product,被Downlevel L3因为我不太了解一个Product,被问到如何给solution时说的太笼统。复习的话其实Meta的DSA内容如果可以随心所欲的掌握,其实这个面试就是个Easy难度。LZ因为只剩下3个Feedback不够用strong所以无法L4送HC,L3暂时没开放HC,蚌埠住了。因为有很多strong所以没有锁定期,3个月后如果还有open面试可以再补一些strong的Feedback可以L4假如L3有Open直接送L3。(挺难受的,总体来说recuriter还算给力)

Amazon

回馈地里加攒人品,大家可以留言问问题,我看见都尽量回

背景: 本人奔4大妈一枚,满打满算5年左右工作经验,中间蛮多GAP,英语比较烂。误打误撞进去FAANG做contract+vendor 做了一路

十一月初被recruiter 在linkedin 上勾搭,抱着试试看的态度去面了🍌 厂

11/18/2021: 第一轮

Business acume + 2 LP (ownership+dive deep)

其实就是BQ 轮,一路狂问问题,无数follow up

12/6/2021: 第二轮

tech 轮, SQL+ customer obssession

SQL 是关于类似物流仓库的表格,有输入时间, 输出时间, 问说当时的库存之类的问题。 总体不难, 最多利口中等水平。window function+group by 基本搞定

follow up 问了有没有其他的metrics你觉得可以算出来,可以提供更多的business insight. 还有当场设计一个dashboard, 你会用怎么设计, 哪些sheet, filter parameter怎么设计, 谁是你的audiens,根据不同的观众,你做出什么相应的调整

1/28/2022: 第三轮 (中间长假,面试官和我都去嗨了,一直推一直推,就到了一月底了)

1. SR. PM Business acume (dive deep+think big)

2. SDE 3 SQL+ high standard : SQL 其实没有真正写, 面试官甩出一堆fields, 问你怎么设计dashboard. fields 是关于类似uber eat 的过程, 有地点,饭店,饭店准备时间,饭店target准备时间,外卖小哥target time,外卖小哥pick up time,target 送达时间, 真正送达时间之类的。问你如何设计dashboard,为什么用柱状图, 为什么用线性图。怎么优化你的dashboard,filter和parameter怎么用。launch之前要注意啥,之类的。

3. bar raiser (B IE): earn trust+ dive deep, millions of follow up questions

4. HM : backbone+ deliver result

5. Sr. BIE: SQL  很简答的SQL(利口简单最多中等), 加上基础概念比如diff between delete and truncate . 还问了tableau 哪些功能我特别喜欢

LP: learn and be curious

2/1 : Offer

总体来说体验不错,recruiter非常奈斯,之前因为出去玩推了final,她也很耐心的帮忙推后。onsite pre也很认真的告诉我注意的点,交流蛮顺畅的。-baidu 1point3acres

先写这些,之后再来更新我觉得面试准备重要的地方

刚面完Phone Screen,来分享下面经攒攒人品。总体体验还是不错的,面试官是个华人小哥,人非常nice。前面coding花了蛮多时间的,所以后面问问题基本上就快问快答了。

SQL 部分:

[hide=188]

给一个transaction, 有 date, order_id , asin, quantity, unit price. 另一个service type table, asin, name( AMZN, FP3, etc..). 要求找到‘2020-02-10’这天number of orders that are exclusively sold by AMZN.

其实非常简单,但楼主上了一天班,脑子卡壳了,在小哥的帮助下还是写全了

Python 部分:

implement logistic regression in python . 1point3acres

非常有意思,目前楼主已经遇到各种要求implement的model了,从tree到kmeans 到LR。不知道小伙伴们有没有相同情况。 Implement 途中小哥还给了一些hint,比如beta0 忘记写进去了之类的。写到gradient descent的时候,要求implement出来, 但是楼主完全忘了log likelyhood 的gradient的公式了,所以就用了pseudo func 代替一下gradient descent。没想到本科数学的推导会在面试出现,大意了啊。-baidu 1point3acres

Stats& ML

1. What is loss function for logistic regress

2. Overfitting vs underfitting

3. Pvalue and confidence interval  

4. Credible interval V.S. confidence interval - credible interval 真是知识盲区了,直接回答不晓得

5. Hypothesis is A/B has significant difference, but the result shows insignificant. How to explain to senior leadership?

6.  Random Forest vs Gradient Boosting

[/hide = 188]

问的问题都偏模型,整体来说难度中等。code 和 A/B testing 的知识没有答道最好。分享一下,以供大家参考

Microsoft

timeline:

11.1海投-11.10收到电面邀请-11.30店面-12.20通知店面通过-1.20onsite面试-1.26offer

和其他面经不一样,我在Onsite之前没有收到选组preference

面试当天连面四轮,每轮一小时,中间没有休息时间,面试官分别是data scientist, principal data scientist,最后一轮是manager面

面试形式:

具体问题就不透露了,基本每一轮都是一半时间深挖简历,提问非常细致,需要对项目非常了解 涉及到DL,NLP相关的知识

然后就是ML DL知识点考察

其次是考coding,我考的是偏算法的coding非SQL,但难度比较低

最后是ML case study,会需要非常详细的设计和detail

我面的组是E&D下面的MLX组,面我的印度小哥人和白人小哥都很好,非常supportive

我投的岗位是data&applied scientist intern,面试通过后transfer为data scientist intern

应该是会接offer啦,欢迎同样收到offer的朋友们勾搭!

Frame: SQL easy+Python medium+Probability+Maths 各一题 共1h.

-----

1. sql easy

给了clickID, pageID 2张表,问calculate CTR

select query, count(distinct clickID)/count(distinct pageID)

from t1 outer join t2

on pageID

group by 1

-baidu 1point3acres

我一开始写的join, follow up 问一般join 默认的inner join, 改成了outer join, pass.

----. check 1point3acres for more.

2. Python medium

get max k value of sliding window in an array

这部分我老老实实说没有python coding exp, 只有data manupulation exp.

但我walk thru我的思路,保持了沟通 面试官会帮助你的思路。

----

3. Probability

15% family 0 child

20% family 1 child

15% family 2 child

20% family 3 child

30% family 4 child

assuming female 0.5

randomly select a female child from the group, prob of the female child has at least one sister?

这里注意一下既然选择了一个child, 那么population是0.85不是1

----

4. 100! vs 10^100 谁大

有点卡 面试官提示一半看

我按照提示懵懵懂懂写完了 还是有点迷糊.

hmmm 我面试中间apt maintenance还咚咚咚敲门送灭火器,尴尬 哈哈哈哈

面试官有引导我思路,都蛮chill的,anyway, 积攒经验吧 道阻且长~

求给点大米 大家要加油呀~ 谢谢大家~

CVS

1. SQL:

a. Get all the members from two table: Union two tables

b. Calculate average score: inner join then avg group by

c. Calculate the proportion of member in each category (gender/zipcode):  group by count() / subquery count total

Intern的面试不要求live coding, 只需要阐述思路。

2. What does randomized experiment do?

3. How to know if the samples are random?

4. What to do if finding out the samples are not random

5. What is overfitting? How to deal with them?

6. How to determine what model to use? What are some business factors you would consider?

7. MC train, a company, wants to build a model to predict whether a customer would cancel

a trip? How would you approach the problem?

IBM

IBM DS Intern OA 数据点来了 大家加油!

180分钟一道题地里原题Predict Mobile App Popularity,没有SQL题目。做之前可以看一眼sample test先熟悉一下环境。Hackerrank平台的notebook里load好了train & test data,要在test data上做prediction。虽然它会自动保存,但是如果像我一样容易手滑还是时不时手动保存一下比较好。最后提交的文件格式一定要对!

train有1900多行,虽然feature不多,但是data cleaning部分还挺麻烦的,花了挺长时间,导致我后面modeling没什么时间好好调参数哭。有很多categorical feature, 还要把一些object类的转成数值,以及处理timestamp. NA倒是不多,全部drop掉也没几个,但可能还是fill更好。Train data挺inbalanced的,本来想直接调包做resampling结果发现sklearn的版本太低了不兼容。。。Modeling之后还要对20个最重要的feature做visualization,不太熟的话确实做起来有点慢。

Netflix

45分钟6题SQL。题目不难,但是test case都很难。无法全部past test case导致楼主没做完。

难的原因是efficiency,10秒内跑完楼主根本没准备这个。

考点有:

join

distinct count

rank

比较两个column算差值(楼主写了半天都是error)

比较三列column哪个最大

. 1point3acres

挂很惨,还是没准备充分。希望有帮助,地里奈飞的面筋好少求加米。

Lyft

一月份海投的粉车,二月初被HR联系面DS, Algorithm and Machine Learning internship。

HR给了三个面试方向:概率统计,优化,ML,我选了ML。邮件里说面试可能会考察classification and regression methods的detail,但实际考的是一个case study: 如何predict whether a user will churn or not. 给的数据是一个表格,里面有user id, driver id, ride starting/end time, prime rate, fare, rating, 等等。感觉主要就是训练一个classification model,但是细节很多,比如缺少数据等等。面试官的问题基本都回答出来了,但是感觉沟通不是特别顺畅,有的idea需要反复clarify,面试后1个工作日通知挂了。

Expedia

1/31 收到OA邀请,大概是一周前海投的简历。. 1point3acres

新人第一次发面经,还在根据草稿回忆选择题中,求加大米,给点动力。

OA 一共16题,前三道是BQ,1分钟时间准备,2分钟答题,可以录两次,如果选录第二次,只会保留最后一遍视频上传。

1.What is DSA in your understanding and why do you want to be this position?2.Why would you like to work for Expedia Group?

3.What strengths can you bring to this role?

4-16是stats选择题(三道大题,每题4-5个小问,每小问2分钟时间)

以下内容需要积分高于 188 您已经可以浏览

[hide=188]

 

 

. From 1point 3acres bbs

1.log in的人中有多少pencentage of booking. check 1point3acres for more.

2.log in和book的人占所有人的多少pencentage

 

 

1.题目中没出现的航空占比多少

2.YA有多少航班delay

 

 

 

1.有多少航班延误时间超过35分钟

2.有多少航班延误时间在25-35分钟之间

 

 

 

OA提交后一个事先录制好的视频介绍了final interview会有三轮,涉及case study, problem solving和bq,还是很紧张的,希望大家都能成功到下一步呀!

Spotify

我看地里spotify的面经很少,所以就分享一下!其实跟之前的面经还是很相似的,主要分成三部分:python, sql, stats

Python的部分就是用numpy和pandas

1. 如何create一个新的column,fill column with conditional values (use np.where).

2. 如何把让第一题里面的conditions flexible to changes (写一个function,然后np.vectorize)

3. 一个很random的follow up,我有点没听懂(欧洲公司口音都令人伤心),大概是如何让上面那个function更flexible,于是我用了dictionary

SQL的部分就是我挂的部分... 也怪lz自己没好好再多刷刷题了,建议各位面试前一天还是多写写SQL练练手感

这道题就是地里之前的题,有两个表,一个是ab_buckets,ab_buckets里面有user_id, ab_buckets, first_expose_date。一个是streams, 里面有user_id, date, category, streamtime。-baidu 1point3acres

1. 这两个表能够告诉我们些什么信息(就是standard的AB test能告诉我们的)

2. 选择什么success metric,如何算出这个metric

a. 注意要用left join,我们不需要不在ab_buckets里面的user

b. 注意要filter出streams的date在expose date之后的

c. 注意b中的filter要考虑到有些user根本不会出现在第二个表中

stats的部分就是比较standard的,什么是p value,什么是statistical power之类的。

希望大家多给我加点米!孩子太穷了已经快要买不起pass了!

Mckinsey

地里的非oa面经好像几乎无,过来贡献一发我的详细经历,tips,和一些case资料,求大米。

OA过了以后邀请到第二轮面试,有3个time slot选择,下周四,周五和下下周五。10/01收到邀请,10/09面试,48 hours出结果,电话通知面试passed。

面试形式:2 interviews back-to-back 顺序不一定

1. Personal experience + Technical expertise 50min

2. Personal experience + case interviews 50 min

具体来说,3 min寒暄+self intro(可有可无),15min personal experience,25min tech/case,剩下时间随便问问题

麦家有个比较贴心的地方是,面试前会举办一场training & Q&A session给所有收到面试邀请的同学,会有官方的讲解面试流程和tips,还有来自前咨询师的ilve demo展示他们是如何比较优秀地完成一个mock case interview的。要吐槽的是他们这个session的时间安排,是和第一个time slot周四的日期重合的,而且我周五面试也莫名没有收到这个training session的邮件邀请,还是我下周面试的朋友告诉我的这个training的内容的。这个安排对第一周面试的同学挺不公平的。

. 1point3acres

McKinsey对于他们面试想考察的点非常开诚布公,在他们官网interview页面也有提到,他们看中candidate‘s 3 characteristics:Inclusive leadership, Entrepreneurship Drive, Personal impact. 对于这三个点的理解可以参考这个网站:https://www.myconsultingoffer.or ... -prep/mckinsey-pei/

每一次Personal experience interview,面试官从3个点里挑一个问你,比如“tell me a time that demonstrate your inclusive leadership in a challenging situation" 。15分钟会dive deep into your selected experience。会问很多follow-up questions,但是都很友好,是真诚地想要知道你当时遇到了什么困难你具体如何解决他们的,不是故意为难你。

Technical expertise在我的面试中没有问特别技术的问题,推模型公式这种都没问(不排除不同的面试官会风格不同,仅作参考)。总体来说就是挑一段你简历上的某个technical project,然后全方面问,你为啥要用这个模型,遇到什么问题,怎么evaluate model success。这个部分我个人答得很好,所有问题都回答上了,而且在面试官跟我讲她的past project的时候,我恰好也对那个领域有一些知识,还和她深入探讨了一下她工作过的project的解决方案。

Case这部分可能是对ds背景同学比较难一些的。我的准备工作是

1. 读了case in point 1-120页,了解了case type and common framework,看完了3个csae讲解。网上零碎看了一些profitability/market entance case资料

2. 自己做了麦肯锡官网的sample case 3个

3. 找人在面试前几个小时mock了两个case(非常的拖延症lol),来源几个MBA program流传的case book

理想情况的话,应该要多mock一些的。。。

case interview的首要tip就是回答一定要MECE(具体你们可以自己查询吧),尽量想全越多的点越好,而且要很有logical structure。第二,计算题要把思路先讲出来,再算。

麦肯锡和其他公司的case不一样,是interviewer-led,意思就是不需要你自己按照framework找面试官要data,面试官直接提供给你了。总共4个sub questions一个一个问下来,像官网提供的sample顺序一样。

我的case面试里,我自己觉得答得不好,因为抽到的industry很陌生,第一题误会了一点面试官的意思,之后又反应比较慢之类的。收到面试通过的结果其实我还蛮悬蛮惊讶的,可能主要还是我的技术面面得很好,中和了case的不足吧。最后,面试官对我真的太友好了,电话告诉我结果的时候还夸我说我虽然回答得不是completely MECE,优点是covered 90% of points in their ideal answer,然后给我进入final round的建议。嗯,总之来自我的经验大概就是,大家不要对case太紧张,不要死背framework,因为大多数时候是套不进去的...

最后,附件内容包括:Case in point, case book(Wharton的有额外几个mckinsey case)

Meta

VO我是2022年一月份面的,是改版后的形式,一共四轮,Analytical Reasoning,Analytical Execution (stats),SQL,Behaviorl.我是4.5年工作经验,就是想冲一下5,HR说这个新形式4和5都可,还是要看我面的怎么样

. check 1point3acres for more.

分了两轮

第一天Analytical Reasoning & Analytical Execution.

Analytical Reasoning: Restaurant Recommendations

FB在考虑build一个餐厅推荐system,插入到user的news feed里面

1. How would you decide if this might be worth while?大概就是问opportunity sizing,要pull什么data之类的

2. How would you design the first iteration of the model?

    1. 我回答logistical model 然后input可以用user的activity history,location,他们friends的activity

    2. 如果没有这些data,可以先推荐local popular restaurants

3. How would you validate your model is working?

    1. 我说可以用A/B test然后看how our key metrics change in the two groups

    2. 另外可以自己抽样,看看recommendation是不是make sense,我们是不是落了什么factor

4. What would you do if ads revenue from restaurants increase 5% but engagement down 3%?

    1. 我就是说先确定这两个是不是有联系,再segment到不同的region和demographic看有没有specific,如果有specific可以看看是哪里出了问题,是不是有cultural difference,如果是的话可以根据那个design一版custom的

5. When would you decide the time to ingest into newsfeed?. From 1point 3acres bbs

    1. 我说是可以看user有没有固定时间每天用这个app的

    2. 另外可以看这个user他如果有很多要看的post,那就先prioritize post;如果他本来每天就能看完,那就可以prioirtize推荐餐厅

6. 你对于上面说的这些所有的还有什么要补充的吗?

我自己反思就是答得还行,但是答得太快了(我语速本来就快,一紧张更快),我的面试官一直在让我慢点说这样他可以记笔记。。然后感觉出了第六个问题就是因为我答得太快了,很快就把人家问题答完了还剩下十五分钟,他就回去看笔记看看还有没有什么可以再问的。。这里其实还挺危险的,所以建议大家还是把握好速度。。

Analytical Execution:

我的面试官直接说是math problem。。然后大概是六七道小题这样的. 1point3acres

前提是Advertiser在fb上买广告,假设target audience size M, purchased N impression

1. Probability an individual see at lease one impression

2. Expected value of total people who see at least one impression

3. We’ve ran a prediction model and discovered 25% of our audience is high intent (90% probability of clicks) and 75% are low intent (10% clicks), how many clicks do we expect to see?

4. If the advertiser are concerned of 0 clicks and want to increase the number of impression they buy. X axis is number of impressions purchased and Y is likelihood of getting 0 clicks, how does X and Y change (draw graph)

    1. 我一开始画的直线,面试官问我所以最后这个probability会是0吗 我说不是,就是接近0

    2. 后来改成应该是 (0.295)^n,n是number of impression,然后as n增加,这个无限趋近于0

5. 如果PM来问你想只target high intent的,你觉得可以吗?

    1. 我怀疑这个题也是凑数的(因为我答的太快了)。。我说是不是可以A/B test然后看看两组表现变化?然后就结束了

第二天

Behavioral:

正常的behavioral题目,包括怎么解决conflict,怎么handle push back,怎么work with people you’ve never worked with before, etc

SQL:

我觉得是一道新题(至少我之前没看到过)

一个表叫虚拟现实,一个叫虚拟现实分类

题目都是问你要解决一个问题,你要pull什么data

比如,我们想知道’game’ category 的user是不是比’home’的要active,你怎么测量?我就自己定义的Number of active days by category这样的。

我感觉因为自己define metrics我基本都往简单的说。。所以题倒是不难

面完之后一个礼拜HR通知过了,给了level 5

整体Timeline

11/15 找朋友内推,提交简历

11/19 跟HR打电话

12/10 第一轮电面

12/15 通知电面过了

1/24-25 VO

1/31 通知过了,接到offer

写了这么多,麻烦大家给加个米!

Tesla

Mechanical Product Design in Crashworthiness Department

  1. There is a car staying before red light in a crossing. Another is coming behind and the brake does not work to reduce the speed. Then, the car behind crashes the car forehead. Which car’s seatbelt receive more stress? Please explain why? (Assume every person in each car is wearing the seatbelt and the seatbelt inside the same car receive the same stress. Also both of the two cars have the same mass and mechanical setting.)

 

  1. There is an ECU (electronic control unit) for air bag reaction system. It keeps collecting data from acceleration and pressure sensor to determine if it necessary to start inflation of airbag. Do you know what is the reaction time or time of delay for ECU when it find the situation to start inflation process? There are 3 options and please explain why?

 

  1. 0.1 seconds

  2. 0.01 seconds

  3. 0.001 seconds

 

Mathworks – Application Engineer (more like control engineering)

  1. Do you know what is the criterion of controllable for controller matrix design?